DOI:https://doi.org/10.65613/737899
Han Jingjing1 Masyitah Abu1*
1.Center for Instructional Technology and Multimedia, Universiti Sains Malaysia
Abstract：In the era of digital learning, music has become a ubiquitous accompanying activity in students’ learning processes. However, existing academic performance prediction models largely overlook the critical dimension of music learning behaviors. This study systematically defines the concept and quantitative indicators of “music-specific cyclic listening behaviors” for the first time. Based on the theory of learning behavioral engagement and the repetitive practice theory of music cognition, two core research hypotheses are proposed. By collecting multi-dimensional learning behavior data and academic performance data from 326 music majors, prediction models incorporating both traditional general indicators and music-specific cyclic listening behavior indicators are constructed. The results show that music-specific cyclic listening behaviors have a significant independent predictive effect on music academic performance (β=0.312, p<0.001). Prediction models incorporating music-specific cyclic listening behavior indicators significantly outperform traditional general indicator models in key evaluation metrics including R², MAE, and RMSE, with R² increasing by 18.7% and MAE decreasing by 15.3%. This study not only expands the feature space of academic performance prediction but also provides new theoretical basis and practical pathways for personalized learning intervention in music education.
Keywords: music-specific cyclic listening behaviors; academic performance prediction; learning behavioral engagement; music cognition; machine learning
- Introduction
1.1 Research Background
With the popularization of digital music platforms, music listening has become an indispensable part of contemporary students’ daily lives and learning. Studies show that over 73% of students have the habit of listening to music while studying, and more than 50% of them believe that music helps their learning. In the field of music education, listening behavior is one of the core links of music learning. Students perceive melodies, understand harmonies, and experience emotions by repeatedly listening to musical works, thereby improving their musical literacy and performance skills.
Meanwhile, the rapid development of educational data mining technology has made it possible to predict students’ academic performance based on learning behavior data. Existing studies have demonstrated that general learning behavior indicators such as class attendance rate, homework completion rate, online learning duration, and forum interaction times can predict students’ academic performance to a certain extent. However, these general indicators are mostly applicable to all disciplines and fail to fully reflect the learning characteristics and behavioral differences of different subjects. Particularly in music education, the relationship between music listening as a discipline-specific learning behavior and academic performance has not been systematically and deeply studied, nor has it been incorporated into academic performance prediction models.
1.2 Problem Statement
Despite the fundamental position of music listening in music learning, existing studies have the following three key problems:
First, there is a lack of clear definition and scientific quantification of “music-specific cyclic listening behaviors”. Most existing studies generally discuss the relationship between “listening to music” and learning, failing to distinguish general background music listening from the specific, purposeful cyclic listening behaviors in the music learning process.
Second, the internal mechanism between music-specific cyclic listening behaviors and music academic performance has not been clarified. Although some studies have shown that repeated listening contributes to music memory and skill acquisition, there is still a lack of empirical evidence on whether this relationship is statistically significant and the extent of its influence.
Third, existing music academic performance prediction models mainly rely on traditional indicators such as classroom performance and examination scores, failing to fully utilize the massive listening behavior data generated by digital music platforms, resulting in limited prediction accuracy and difficulty in meeting the needs of personalized music education.
Based on the above problems, this study aims to explore: Can music-specific cyclic listening behaviors independently predict music academic performance? Can incorporating music-specific cyclic listening behavior indicators into prediction models significantly improve model prediction accuracy?
1.3 Research Significance
1.3.1 Theoretical Significance
This study systematically defines the concept of music-specific cyclic listening behaviors for the first time and constructs its quantitative indicator system, enriching the research content of music learning behavior theory. Meanwhile, this study applies the theory of learning behavioral engagement and the repetitive practice theory of music cognition to the field of music academic performance prediction, revealing the internal mechanism by which music-specific cyclic listening behaviors affect academic performance, and providing a new theoretical perspective for understanding the music learning process.
1.3.2 Practical Significance
The results of this study can provide scientific decision-making basis for music educators. By analyzing students’ cyclic listening behavior data, teachers can timely identify students with learning difficulties and carry out targeted intervention and guidance. In addition, the high-precision prediction model constructed in this study can be integrated into online music learning platforms to provide students with personalized learning resource recommendations and learning path planning, improving the quality and efficiency of music education.
1.4 Research Innovations
The innovations of this study are mainly reflected in the following three aspects:
First, conceptual innovation. This study clearly distinguishes general background music listening from specific cyclic listening behaviors in the music learning process for the first time, and constructs a quantitative indicator system of music-specific cyclic listening behaviors from four dimensions: behavior frequency, behavior duration, behavior depth, and behavior pattern.
Second, methodological innovation. This study combines objective listening behavior data generated by digital music platforms with traditional learning behavior data, adopts multiple machine learning algorithms to construct prediction models, and verifies the improvement effect of music-specific cyclic listening behavior indicators on prediction accuracy through comparative experiments.
Third, perspective innovation. Starting from the perspective of discipline-specific learning behaviors, this study explores the prediction of music academic performance, breaking through the limitation of existing studies that mainly rely on general learning behavior indicators, and providing reference for academic performance prediction research in other disciplines.
1.5 Research Content and Technical Route
The main contents of this study include:
(1)Literature review and core research gap demonstration: Systematically sort out domestic and foreign research literature on music listening behaviors, learning behavioral engagement, academic performance prediction, etc., and clarify the deficiencies of existing studies and the entry point of this study.
(2)Theoretical basis and research hypotheses: Based on the theory of learning behavioral engagement and the repetitive practice theory of music cognition, propose two core hypotheses of this study.
(3)Research design and methods: Determine research subjects, design data collection schemes, construct prediction models, and select appropriate model evaluation indicators.
(4)Research results: Present the results of descriptive statistical analysis, correlation analysis, regression analysis, and model comparison experiments.
(5)Discussion: Conduct in-depth analysis and interpretation of the research results, and discuss their theoretical and practical significance.
(6)Conclusion: Summarize the main findings of the study and clarify the contributions of the research.
The technical route of this study is as follows: First, conduct literature research and theoretical analysis to propose research hypotheses; then design a data collection scheme to collect students’ learning behavior data and academic performance data; next preprocess the data and perform feature engineering; then construct traditional general indicator models and improved models incorporating music-specific cyclic listening behavior indicators; finally compare and evaluate the performance of the two models to verify the research hypotheses.
- Literature Review and Core Research Gap Demonstration
2.1 Research on the Relationship Between Music Listening Behaviors and Learning
The relationship between music and learning has long been a research hotspot in psychology and education. Early “Mozart effect” studies found that listening to Mozart’s music could temporarily improve individuals’ spatial reasoning ability. However, subsequent research results have been inconsistent, with some studies supporting the promoting effect of music on learning and others finding that music interferes with learning. LUO Can integrated 127 relevant studies through meta-analysis and found that there are significant moderating effects in the impact of music on learning, among which music type, learning task type, and individual differences are the most important moderating variables[1].
This inconsistent result is mainly due to the failure of studies to distinguish different types of music listening behaviors. Most existing studies focus on the impact of background music on general learning tasks, ignoring the specific listening behaviors in the music learning process. WANG Yunlu found through controlled experiments that the attention stability score of college students under the single-track loop condition was significantly higher than that under the random play condition, indicating that purposeful and repeated music listening may have different cognitive effects from general background music listening[4]. Wang et al. further pointed out that music can promote learning when it is semantically related to the learning task; otherwise, it may cause interference[2].
In the field of music education, listening is considered the foundation of music learning. Students establish musical concepts, develop musical perception abilities, and understand the structure and emotion of music by listening to musical works. However, most existing studies remain at the qualitative description level, lacking quantitative analysis of music listening behaviors, and even less exploring the quantitative relationship between them and academic performance. Hassan found a positive correlation between music listening duration and music academic performance through questionnaire survey, but this study did not distinguish the difference between cyclic listening and random listening, nor did it control the influence of other confounding variables[6].
2.2 Theory of Learning Behavioral Engagement and Its Application
The theory of learning behavioral engagement was proposed by Fredricks et al., which divides learning engagement into three dimensions: behavioral engagement, emotional engagement, and cognitive engagement[9]. Among them, behavioral engagement refers to the degree to which learners participate in academic, social, and extracurricular activities, including observable behavioral manifestations such as class attendance rate, homework completion rate, and learning duration. The theory holds that these three dimensions are interrelated and interact with each other, jointly determining students’ learning effects.
The theory of learning behavioral engagement has been widely applied in academic performance prediction research. A large number of studies have shown that behavioral engagement is an important predictor of academic performance, and the higher the degree of students’ behavioral engagement, the better their academic performance tends to be. ZHANG Xue and LI Ming found through meta-analysis that the average effect size of behavioral engagement on academic performance was 0.42, significantly higher than the effect sizes of emotional engagement and cognitive engagement[8].
In music education, learning behavioral engagement includes not only general behaviors such as classroom participation and homework completion but also discipline-specific behaviors such as music listening, instrument practice, and music creation. Mahmood et al. found through structural equation modeling that behavioral engagement in music learning has a significant positive impact on students’ music academic performance, with an effect size of 0.51[7]. However, this study did not analyze music listening behavior as an independent dimension, but merged it with other learning behaviors into a comprehensive indicator, which to some extent masked the unique role of music listening behavior.
2.3 Repetitive Practice Theory of Music Cognition
Repetitive practice is the basic way of skill acquisition, which is particularly important in music learning. The repetitive practice theory of music cognition holds that repeated listening and performance can strengthen neural connections in the brain, form stable music memory and muscle memory, thereby improving the proficiency of music skills. Levitin found through neuroimaging studies that there are stronger neural connections between the auditory cortex and motor cortex of professional musicians, which is the result of long-term repetitive practice[11].
Neuroimaging studies have shown that repetitive practice can activate brain regions related to learning and memory formation, including the hippocampus, cerebellum, and temporal lobe. When students repeatedly listen to the same musical work, their brains gradually become familiar with the melody, rhythm, harmony, and structure of the work, thus being able to understand the connotation of the work more deeply. The “repetition paradox” theory proposed by Margulis points out that although repetition reduces the novelty of stimuli, it also increases individuals’ emotional involvement and depth of understanding of stimuli, which is particularly evident in music listening[14].
However, most existing studies on music repetitive practice focus on instrument performance practice, while relatively few studies focus on music listening practice. Although some studies have shown that repeated listening contributes to music memory, there is still a lack of empirical evidence on whether this memory advantage can be transformed into academic performance improvement. LIU Chang pointed out that there is a distinction between “effective repetition” and “ineffective repetition” in piano learning, and only purposeful and feedback-oriented repetitive practice can truly improve performance skills[13]. This view also applies to music listening practice, but there is currently no research that systematically defines and quantifies “effective repetition” in music listening.
2.4 Research on Academic Performance Prediction Models
Academic performance prediction is one of the core research directions in the field of educational data mining. Existing studies have adopted various machine learning algorithms to construct prediction models, including logistic regression, decision trees, random forests, support vector machines, and neural networks. These models usually use students’ demographic characteristics, historical academic performance, learning behavior data, etc. as input features. ZHAN Zhaoman et al. systematically reviewed 156 academic performance prediction studies and found that random forests and gradient boosting trees are currently the two best-performing algorithms[16].
In the field of music education, some studies have also attempted to construct music academic performance prediction models. For example, Li et al. constructed a personalized prediction model for high school students’ music learning effects based on deep learning, with an accuracy rate of 87.6%[17]. Zhang et al. constructed a hybrid architecture model based on XGBoost and BiLSTM, with a prediction accuracy rate of 89.2% in rhythm training scenarios[19]. Chen et al. further applied multimodal data fusion technology to music teaching effect prediction and achieved better prediction results[20].
However, existing music academic performance prediction models mainly rely on traditional indicators such as classroom performance, examination scores, and practice duration, failing to fully utilize the massive listening behavior data generated by digital music platforms. In particular, music-specific cyclic listening behavior, an important learning behavior indicator, has not been incorporated into prediction models, which to some extent limits the prediction accuracy of the models. WANG Hao and LI Na pointed out that future music academic performance prediction research should pay more attention to discipline-specific learning behavior data to further improve the prediction ability of models[18].
2.5 Core Research Gaps
Through systematic sorting of existing literature, the following three core research gaps can be identified:
First, ambiguous conceptual definition. Existing studies have failed to clearly distinguish general background music listening from specific cyclic listening behaviors in the music learning process, nor have they established a scientific quantitative indicator system. Most studies simply use “music listening duration” as the only measurement indicator, which cannot reflect the quality and pattern differences of listening behaviors.
Second, insufficient mechanism research. Although some studies have shown that repeated listening contributes to music learning, the internal mechanism by which music-specific cyclic listening behaviors affect academic performance has not been systematically and deeply explored. In particular, the different impacts of different types of cyclic listening behaviors (such as deep cyclic vs. fragmented cyclic) on academic performance, and the differences of these impacts among students of different majors and grades, still lack empirical evidence.
Third, imperfect prediction models. Existing music academic performance prediction models mainly rely on traditional general indicators, failing to fully utilize music-specific listening behavior data, resulting in limited prediction accuracy. Meanwhile, most existing studies only focus on the overall performance of the model, and rarely discuss the contribution degree of different features to the model prediction results, which makes the model less interpretable and difficult to apply to actual educational teaching practice.
This study is carried out precisely in response to the above research gaps, aiming to reveal the relationship between music-specific cyclic listening behaviors and music academic performance through empirical research, and verify their role in improving the accuracy of academic performance prediction.
- Theoretical Basis and Research Hypotheses
3.1 Theory of Learning Behavioral Engagement
The theory of learning behavioral engagement holds that students’ learning engagement is a multi-dimensional concept, including three interrelated dimensions: behavioral engagement, emotional engagement, and cognitive engagement. Among them, behavioral engagement is the most basic and easily observable dimension, which reflects the degree and quality of students’ participation in learning activities.
In music education, behavioral engagement includes not only general behaviors such as class attendance, homework completion, and classroom interaction but also discipline-specific behaviors such as music listening, instrument practice, and music creation. Music-specific cyclic listening behavior, as an important music learning behavior, is a concrete manifestation of students’ music learning behavioral engagement. Unlike passive behaviors such as class attendance, cyclic listening behavior more reflects students’ active learning willingness and self-regulation ability.
According to the theory of learning behavioral engagement, the higher the degree of students’ behavioral engagement, the better their academic performance tends to be. This is because higher behavioral engagement means that students spend more time and energy on learning and participate in more learning activities, thus being able to acquire more knowledge and skills. Therefore, we can infer that music-specific cyclic listening behavior, as an important dimension of music learning behavioral engagement, should have a significant positive predictive effect on music academic performance.
3.2 Repetitive Practice Theory of Music Cognition
The repetitive practice theory of music cognition holds that the acquisition of music skills is a process of continuously strengthening neural connections through repeated practice. When students repeatedly listen to the same musical work, their brains gradually form neural representations of the work, including various aspects such as melody, rhythm, harmony, and structure.
The formation of such neural representations has the following effects: First, it can improve students’ music perception ability, enabling them to capture details and changes in music more sensitively; second, it can enhance students’ music memory ability, enabling them to recall and reproduce musical works more accurately; third, it can deepen students’ understanding of musical works, enabling them to better experience the emotion and connotation of the works.
All these improvements will eventually be reflected in students’ music academic performance. For example, in sight-singing and ear-training examinations, students with rich cyclic listening experience can identify pitch and rhythm more accurately; in music appreciation examinations, they can analyze and evaluate musical works more deeply; in instrument performance examinations, they can grasp the style and emotion of the works more accurately. Therefore, according to the repetitive practice theory of music cognition, music-specific cyclic listening behaviors should have a significant positive impact on music academic performance.
3.3 Research Hypotheses
Based on the above theoretical analysis, this study proposes the following two core research hypotheses:
H1: Music-specific cyclic listening behaviors have a significant independent predictive effect on music academic performance
This hypothesis holds that after controlling for traditional general learning behavior indicators (such as class attendance rate, homework completion rate, practice duration, etc.), music-specific cyclic listening behaviors can still significantly predict music academic performance. This means that music-specific cyclic listening behavior is an independent factor affecting music academic performance, not just a by-product of other learning behaviors.
H2: Prediction models incorporating music-specific cyclic listening behavior indicators have significantly better accuracy than traditional general indicator models
This hypothesis holds that after incorporating music-specific cyclic listening behavior indicators into the prediction model, the prediction accuracy of the model will be significantly improved. This is because music-specific cyclic listening behavior indicators can capture important information in the music learning process that cannot be reflected by traditional general indicators, thus enabling the model to predict students’ academic performance more accurately.
- Research Design and Methods
4.1 Research Subjects
This study adopted a cluster sampling method and selected 326 undergraduate students majoring in music from 3 institutions of higher education as research subjects. Among them, 128 were male and 198 were female; 87 were freshmen, 92 were sophomores, 85 were juniors, and 62 were seniors; 105 majored in piano, 98 in vocal music, 76 in instrumental music, and 47 in music theory.
All research subjects used the same online music learning platform for music learning, which could automatically record all their listening behavior data. The age of the research subjects ranged from 18 to 24 years, with an average age of 20.3±1.5 years. All research subjects voluntarily participated in this study and signed informed consent forms.
4.2 Data Collection and Preprocessing
4.2.1 Data Collection
The data collected in this study include three parts: music-specific cyclic listening behavior data, traditional general learning behavior data, and music academic performance data.
Music-specific cyclic listening behavior data: Automatically collected through the online music learning platform, with a collection period of 5 months from September 2024 to January 2025. The collected indicators include:
- Total number of cyclic listening: The total number of times students cyclically played musical works during the learning process
- Total duration of cyclic listening: The total duration of students’ cyclic listening during the learning process (minutes)
- Average number of cycles: The average number of cyclic plays per musical work by students
- Deep cyclic rate: The proportion of musical works that students cyclically played ≥10 times
- Fragmented cyclic rate: The proportion of students’ single cyclic listening duration <5 minutes
Traditional general learning behavior data: Collected through the school academic affairs system and the online music learning platform, including:
- Class attendance rate: The ratio of students’ actual attendance times to the required attendance times
- Homework completion rate: The ratio of students’ on-time homework completion times to the total homework times
- Online learning duration: The total learning duration of students on the online music learning platform (minutes)
- Forum interaction times: The total number of posts and replies by students on the learning forum
- Instrument practice duration: The average weekly instrument practice duration self-reported by students (hours)
Music academic performance data: Collected through the school academic affairs system, including students’ final examination scores of all music courses in the first semester of the 2024-2025 academic year. This study uses the average score of all music courses as the final academic performance indicator, which can more comprehensively reflect students’ overall music learning level.
4.2.2 Data Preprocessing
Before data analysis, the collected data were preprocessed first, including missing value processing, outlier processing, and data standardization.
Missing value processing: For a small amount of missing data (<5%), the mean imputation method was used for processing. For data with more missing values (≥5%), the sample was deleted from the dataset. Finally, a total of 12 samples were deleted, resulting in 314 valid samples.
Outlier processing: The Z-score method was used to identify outliers. For data with absolute Z-score greater than 3, they were regarded as outliers and processed by the median replacement method.
Data standardization: Due to the different dimensions and value ranges of different indicators, in order to eliminate the influence of dimensions, the Z-score standardization method was used to standardize all continuous variables, so that the mean of each variable was 0 and the standard deviation was 1.
4.3 Prediction Model Construction
This study adopted multiple machine learning algorithms to construct prediction models and selected the algorithm with the best performance for the final model comparison. The algorithms selected in this study include: Linear Regression, Support Vector Regression (SVR), Random Forest Regression (RFR), and Gradient Boosting Regression (GBR).
4.3.1 Feature Selection
This study constructed two feature sets:
- Traditional general feature set: Contains 5 features: class attendance rate, homework completion rate, online learning duration, forum interaction times, and instrument practice duration.
- Fusion feature set: Based on the traditional general feature set, adds 5 music-specific cyclic listening behavior features: total number of cyclic listening, total duration of cyclic listening, average number of cycles, deep cyclic rate, and fragmented cyclic rate, totaling 10 features.
To avoid feature redundancy, the Pearson correlation coefficient method was used to analyze the correlation between features. The results showed that the correlation coefficients between all features were less than 0.7, indicating that there was no serious multicollinearity problem and all could be included in the model.
4.3.2 Model Training and Validation
The preprocessed dataset was randomly divided into a training set and a test set according to a 7:3 ratio. Among them, the training set contained 220 samples for model training, and the test set contained 94 samples for model performance evaluation.
A 5-fold cross-validation method was used to train and tune the models. For each algorithm, the optimal hyperparameter combination was found through grid search. All models were implemented using the scikit-learn library in the Python 3.9 environment.
4.4 Model Evaluation Indicators
This study adopted the following four commonly used regression model evaluation indicators to evaluate the performance of the models:
(1)Coefficient of determination (R²): Indicates the proportion of dependent variable variation that the model can explain, with a value range of 0 to 1. The larger the value, the better the fitting effect of the model.
(2)Mean Absolute Error (MAE): Indicates the average of the absolute errors between the predicted values and the true values. The smaller the value, the higher the prediction accuracy of the model.
(3)Root Mean Square Error (RMSE): Indicates the square root of the mean of the squared errors between the predicted values and the true values. The smaller the value, the higher the prediction accuracy of the model.
(4)Mean Absolute Percentage Error (MAPE): Indicates the average of the percentage of absolute errors between the predicted values and the true values. The smaller the value, the higher the prediction accuracy of the model.
- Research Results
5.1 Descriptive Statistical Analysis
Table 1 shows the overall descriptive statistical results of all variables. As can be seen from the table, the average music academic score of students was 78.5±9.2, indicating a good overall learning level. In terms of music-specific cyclic listening behaviors, the average total number of cyclic listening of students was 128.6±65.3 times, the average total duration of cyclic listening was 452.3±215.7 minutes, the average number of cycles was 4.2±2.1 times, the deep cyclic rate was 0.23±0.15, and the fragmented cyclic rate was 0.35±0.18.
Table 1 Overall descriptive statistical results of variables
| Variable | Mean | Standard Deviation | Minimum | Maximum |
| Average music academic score | 78.5 | 9.2 | 52.0 | 96.0 |
| Class attendance rate | 0.92 | 0.08 | 0.65 | 1.00 |
| Homework completion rate | 0.89 | 0.11 | 0.58 | 1.00 |
| Online learning duration (minutes) | 876.5 | 342.1 | 215.0 | 1892.0 |
| Forum interaction times | 12.3 | 8.7 | 0.0 | 45.0 |
| Instrument practice duration (hours/week) | 12.5 | 5.8 | 3.0 | 28.0 |
| Total number of cyclic listening | 128.6 | 65.3 | 23.0 | 356.0 |
| Total duration of cyclic listening (minutes) | 452.3 | 215.7 | 87.0 | 1245.0 |
| Average number of cycles | 4.2 | 2.1 | 1.0 | 12.5 |
| Deep cyclic rate | 0.23 | 0.15 | 0.00 | 0.72 |
| Fragmented cyclic rate | 0.35 | 0.18 | 0.08 | 0.82 |
To further understand the differences in cyclic listening behaviors among students of different majors, this study conducted one-way ANOVA on the cyclic listening behavior indicators of the four majors, and the results are shown in Table 2.
Table 2 Comparison of cyclic listening behavior indicators among students of different majors
| Major | Sample Size | Total number of cyclic listening | Total duration of cyclic listening (minutes) | Average number of cycles | Deep cyclic rate | Fragmented cyclic rate |
| Piano | 101 | 142.3±72.5 | 512.6±234.8 | 4.8±2.3 | 0.27±0.16 | 0.32±0.17 |
| Vocal music | 95 | 135.7±68.2 | 478.3±221.5 | 4.5±2.2 | 0.25±0.15 | 0.34±0.18 |
| Instrumental music | 73 | 118.5±59.6 | 412.5±198.7 | 3.9±2.0 | 0.21±0.14 | 0.36±0.19 |
| Music theory | 45 | 102.4±52.3 | 356.2±176.4 | 3.5±1.8 | 0.18±0.12 | 0.39±0.20 |
| F value | – | 4.236 | 5.127 | 3.892 | 4.563 | 2.145 |
| p value | – | 0.006 | 0.002 | 0.009 | 0.004 | 0.094 |
As can be seen from Table 2, there are significant differences in cyclic listening behaviors among students of different majors. Piano majors have significantly higher total number of cyclic listening, total duration, average number of cycles, and deep cyclic rate than other majors, while music theory majors have significantly lower these indicators than other majors. This result is consistent with the learning characteristics of different majors. Piano and vocal music majors need to learn performance/singing skills through more listening and imitation, while music theory majors focus more on theoretical knowledge learning.
5.2 Correlation Analysis
Table 3 shows the Pearson correlation coefficients between all variables. As can be seen from the table, all traditional general learning behavior indicators and music-specific cyclic listening behavior indicators are significantly positively correlated with the average music academic score (p<0.01). Among them, the deep cyclic rate has the highest correlation coefficient with the average music academic score (r=0.425), followed by the total duration of cyclic listening (r=0.398) and the average number of cycles (r=0.376).
Table 3 Pearson correlation coefficients between variables
| Variable | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
| 1. Average music academic score | 1.000 | |||||||||
| 2. Class attendance rate | 0.352 | 1.000 | ||||||||
| 3. Homework completion rate | 0.387 | 0.421 | 1.000 | |||||||
| 4. Online learning duration | 0.315 | 0.368 | 0.392 | 1.000 | ||||||
| 5. Forum interaction times | 0.289 | 0.256 | 0.312 | 0.345 | 1.000 | |||||
| 6. Instrument practice duration | 0.403 | 0.321 | 0.357 | 0.389 | 0.278 | 1.000 | ||||
| 7. Total number of cyclic listening | 0.364 | 0.287 | 0.315 | 0.352 | 0.245 | 0.326 | 1.000 | |||
| 8. Total duration of cyclic listening | 0.398 | 0.302 | 0.334 | 0.376 | 0.267 | 0.351 | 0.875 | 1.000 | ||
| 9. Average number of cycles | 0.376 | 0.275 | 0.308 | 0.341 | 0.239 | 0.318 | 0.789 | 0.823 | 1.000 | |
| 10. Deep cyclic rate | 0.425 | 0.318 | 0.347 | 0.382 | 0.271 | 0.365 | 0.756 | 0.798 | 0.852 | 1.000 |
| 11. Fragmented cyclic rate | -0.187 | -0.125* | -0.142* | -0.158* | -0.112* | -0.163* | -0.215 | -0.238 | -0.207 | -0.256 |
Note: p<0.05, *p<0.01
It is worth noting that the fragmented cyclic rate is significantly negatively correlated with the average music academic score (r=-0.187, p<0.01), indicating that excessive fragmented cyclic listening may be detrimental to music learning. In addition, there is a high correlation between cyclic listening behavior indicators, which is because they all reflect students’ cyclic listening behavior characteristics. However, since the correlation coefficients are all less than 0.9, they will not cause serious multicollinearity problems in regression analysis.
5.3 Hierarchical Regression Analysis
To more accurately verify hypothesis H1, this study adopted the hierarchical regression analysis method and constructed regression models in two steps. In the first step, only traditional general learning behavior indicators were included in the model (Model 1); in the second step, music-specific cyclic listening behavior indicators were added to Model 1 (Model 2). The hierarchical regression analysis results are shown in Table 4.
Table 4 Hierarchical regression analysis results
| Variable | Model 1 | Model 2 | ||||
| B | SE | β | B | SE | β | |
| Constant | 78.523 | 0.521 | – | 78.523 | 0.452 | – |
| Class attendance rate | 4.125 | 1.234 | 0.162 | 3.256 | 1.125 | 0.128 |
| Homework completion rate | 4.567 | 1.156 | 0.184 | 3.872 | 1.058 | 0.156* |
| Online learning duration | 2.876 | 0.965 | 0.131* | 2.145 | 0.876 | 0.098* |
| Forum interaction times | 2.123 | 0.812 | 0.097* | 1.568 | 0.723 | 0.072* |
| Instrument practice duration | 5.678 | 1.345 | 0.235* | 4.523 | 1.235 | 0.187* |
| Total number of cyclic listening | – | – | – | 1.876 | 0.987 | 0.085 |
| Total duration of cyclic listening | – | – | – | 2.345 | 1.052 | 0.108* |
| Average number of cycles | – | – | – | 2.789 | 1.123 | 0.125* |
| Deep cyclic rate | – | – | – | 5.236 | 1.345 | 0.312* |
| Fragmented cyclic rate | – | – | – | -2.156 | 0.987 | -0.098* |
| R² | 0.423 | 0.587 | ||||
| Adjusted R² | 0.410 | 0.573 | ||||
| F value | 32.456* | 41.234* | ||||
| ΔR² | – | 0.164* |
Note: p<0.05, p<0.01, **p<0.001
As can be seen from Table 4, the R² of Model 1 is 0.423, indicating that traditional general learning behavior indicators can explain 42.3% of the variation in music academic performance. After adding music-specific cyclic listening behavior indicators, the R² of Model 2 increased to 0.587, with ΔR² of 0.164, and this increase was statistically significant (p<0.001). This indicates that music-specific cyclic listening behavior indicators can explain an additional 16.4% of the variation in academic performance that cannot be explained by traditional general indicators.
In Model 2, deep cyclic rate (β=0.312, p<0.001), total duration of cyclic listening (β=0.108, p<0.05), average number of cycles (β=0.125, p<0.05), and fragmented cyclic rate (β=-0.098, p<0.05) all have significant predictive effects on the average music academic score. Among them, the standardized coefficient of deep cyclic rate is the largest, indicating that it is the most influential factor on music academic performance among all variables. This result fully verifies hypothesis H1, that is, music-specific cyclic listening behaviors have a significant independent predictive effect on music academic performance.
5.4 Model Comparison Analysis
To verify hypothesis H2, this study used traditional general feature sets and fusion feature sets respectively to construct prediction models using four different machine learning algorithms, and evaluated the performance of the models on the test set. The model comparison results are shown in Table 5.
Table 5 Performance comparison of different models
| Algorithm | Feature Set | R² | MAE | RMSE | MAPE(%) |
| Linear Regression | Traditional general feature set | 0.423 | 5.678 | 7.234 | 7.892 |
| Linear Regression | Fusion feature set | 0.502 | 4.892 | 6.345 | 6.785 |
| SVR | Traditional general feature set | 0.456 | 5.345 | 6.987 | 7.456 |
| SVR | Fusion feature set | 0.538 | 4.567 | 6.012 | 6.342 |
| RFR | Traditional general feature set | 0.512 | 4.987 | 6.543 | 6.987 |
| RFR | Fusion feature set | 0.608 | 4.123 | 5.432 | 5.678 |
| GBR | Traditional general feature set | 0.534 | 4.765 | 6.321 | 6.754 |
| GBR | Fusion feature set | 0.634 | 3.892 | 5.123 | 5.345 |
As can be seen from Table 5, for all four algorithms, the performance of models using the fusion feature set is significantly better than that of models using the traditional general feature set. Among them, the Gradient Boosting Regression (GBR) algorithm has the best performance. The R² of the GBR model using the fusion feature set reaches 0.634, which is 18.7% higher than that of the GBR model using the traditional general feature set; MAE is 3.892, decreased by 15.3%; RMSE is 5.123, decreased by 18.9%; MAPE is 5.345%, decreased by 20.9%.
To further verify whether this performance improvement is statistically significant, this study used a paired t-test to compare the prediction errors of the two GBR models. The results showed that the prediction error of the GBR model using the fusion feature set was significantly smaller than that of the GBR model using the traditional general feature set (t=4.236, p<0.001). This result fully verifies hypothesis H2, that is, prediction models incorporating music-specific cyclic listening behavior indicators have significantly better accuracy than traditional general indicator models.
To further understand the performance differences of different models in different majors, this study evaluated the models on the test samples of the four majors respectively, and the results are shown in Table 6.
Table 6 Model performance comparison of different majors (GBR algorithm)
| Major | Traditional general feature set | Fusion feature set | R² improvement rate (%) | ||
| R² | MAE | R² | MAE | ||
| Piano | 0.512 | 4.876 | 0.658 | 3.765 | 28.5 |
| Vocal music | 0.498 | 5.023 | 0.624 | 3.987 | 25.3 |
| Instrumental music | 0.526 | 4.765 | 0.612 | 4.012 | 16.4 |
| Music theory | 0.543 | 4.654 | 0.597 | 4.234 | 9.9 |
As can be seen from Table 6, the fusion feature set model outperforms the traditional general feature set model in all majors, but there are significant differences in the improvement rate. Among them, the R² improvement rate of piano majors is the largest, reaching 28.5%; followed by vocal music majors, with an improvement rate of 25.3%; music theory majors have the smallest improvement rate, only 9.9%. This result is consistent with the learning characteristics of different majors. The learning of piano and vocal music majors is more dependent on cyclic listening behaviors, so cyclic listening behavior indicators contribute more to the academic performance prediction of these majors.
5.5 Feature Importance Analysis
To further understand the contribution degree of each feature to the model prediction results, this study conducted feature importance analysis on the best-performing fusion feature set GBR model, and the results are shown in Table 7.
Table 7 Feature importance ranking of the fusion feature set GBR model
| Feature | Importance Score | Ranking |
| Deep cyclic rate | 0.235 | 1 |
| Instrument practice duration | 0.187 | 2 |
| Homework completion rate | 0.152 | 3 |
| Total duration of cyclic listening | 0.123 | 4 |
| Average number of cycles | 0.108 | 5 |
| Class attendance rate | 0.087 | 6 |
| Online learning duration | 0.056 | 7 |
| Fragmented cyclic rate | 0.032 | 8 |
| Total number of cyclic listening | 0.018 | 9 |
| Forum interaction times | 0.002 | 10 |
As can be seen from Table 7, deep cyclic rate is the most important feature, with an importance score of 0.235, much higher than other features. Followed by instrument practice duration (0.187), homework completion rate (0.152), total duration of cyclic listening (0.123), and average number of cycles (0.108). This result is consistent with the regression analysis results, further proving that music-specific cyclic listening behaviors, especially deep cyclic behaviors, have an important impact on music academic performance.
To further understand the differences in feature importance among different majors, this study conducted feature importance analysis on the GBR models of the four majors respectively, and the results are shown in Table 8.
Table 8 Feature importance ranking of different majors (top 5)
| Ranking | Piano | Vocal music | Instrumental music | Music theory |
| 1 | Deep cyclic rate (0.287) | Deep cyclic rate (0.265) | Instrument practice duration (0.243) | Homework completion rate (0.215) |
| 2 | Instrument practice duration (0.198) | Instrument practice duration (0.187) | Deep cyclic rate (0.212) | Class attendance rate (0.198) |
| 3 | Total duration of cyclic listening (0.156) | Homework completion rate (0.164) | Homework completion rate (0.176) | Instrument practice duration (0.176) |
| 4 | Homework completion rate (0.132) | Total duration of cyclic listening (0.143) | Class attendance rate (0.134) | Deep cyclic rate (0.152) |
| 5 | Average number of cycles (0.108) | Average number of cycles (0.112) | Total duration of cyclic listening (0.105) | Online learning duration (0.123) |
As can be seen from Table 8, there are significant differences in the feature importance rankings among different majors. For piano and vocal music majors, deep cyclic rate is the most important feature; for instrumental music majors, instrument practice duration is the most important feature; for music theory majors, homework completion rate is the most important feature. This result further proves that there are differences in the learning characteristics and key success factors of different majors, so when constructing academic performance prediction models, the discipline-specific learning behavior indicators should be fully considered.
- Discussion
6.1 Independent Predictive Effect of Music-Specific Cyclic Listening Behaviors on Academic Performance
The results of this study show that music-specific cyclic listening behaviors have a significant independent predictive effect on music academic performance, and this predictive effect still exists even after controlling for traditional general learning behavior indicators. This result is consistent with the expectations of the theory of learning behavioral engagement and the repetitive practice theory of music cognition, and also provides new empirical evidence for the application of these two theories in the field of music education.
According to the theory of learning behavioral engagement, music-specific cyclic listening behavior is a concrete manifestation of students’ music learning behavioral engagement. Unlike passive behaviors such as class attendance, cyclic listening behavior more reflects students’ active learning willingness and self-regulation ability. The more times, longer duration, and deeper depth of cyclic listening students conduct, the higher their behavioral engagement in music learning, and thus the better their academic performance. This study found that deep cyclic rate is the most important factor affecting music academic performance, indicating that purposeful and in-depth cyclic listening behavior can better reflect students’ learning engagement than simple listening duration.
According to the repetitive practice theory of music cognition, repeated listening can strengthen neural connections in the brain, form stable music memory and neural representations, thereby improving students’ music perception ability, memory ability, and understanding ability. In particular, deep cyclic behavior (cyclic playing a single work ≥10 times) enables students to understand the structure, harmony, and emotion of musical works more deeply. Neuroscience studies have shown that when individuals repeatedly listen to the same musical work, the auditory cortex, hippocampus, and prefrontal cortex in the brain gradually form a synergistic activation pattern, which is closely related to the consolidation of music memory and the deepening of understanding[11,14]. The results of this study are consistent with this neural mechanism. Students with higher deep cyclic rates have stronger music memory and understanding abilities, and thus better academic performance.
It is worth noting that the fragmented cyclic rate is significantly negatively correlated with music academic performance. This may be because excessive fragmented cyclic listening indicates that students’ learning attention is not concentrated and their learning lacks systematicity and continuity. Cognitive psychology studies have shown that fragmented learning disrupts the integration process of working memory, leading to fragmented knowledge and superficial understanding[12]. In music learning, fragmented cyclic listening may cause students to only focus on certain fragments of musical works, unable to form a complete understanding of the overall structure and emotion of the works. This result suggests that in music learning, we should not only pay attention to the quantity of cyclic listening but also to the quality of cyclic listening, and guide students to conduct purposeful and systematic in-depth listening.
This study also found that there are significant differences in cyclic listening behaviors among students of different majors, and the degree of influence of cyclic listening behaviors on academic performance also varies by major. Piano and vocal music majors have significantly higher total number of cyclic listening, total duration, and deep cyclic rate than other majors, and cyclic listening behavior indicators contribute the most to the academic performance prediction of these majors. This is because the learning of piano and vocal music majors relies more on auditory imitation and perception, and students need to learn performance/singing skills and grasp the style and emotion of works through repeated listening. In contrast, the learning of instrumental music majors focuses more on the training of instrumental performance skills, so the duration of instrument practice has a greater impact on academic performance. The learning of music theory majors focuses more on the mastery of theoretical knowledge, so homework completion rate and class attendance rate have a greater impact on academic performance. This result indicates that music learning behaviors have significant disciplinary specificity, and the learning characteristics of different majors should be fully considered in music education research and practice.
6.2 Improvement Effect of Music-Specific Cyclic Listening Behaviors on Prediction Accuracy
The results of this study also show that prediction models incorporating music-specific cyclic listening behavior indicators have significantly better accuracy than traditional general indicator models. This result has important theoretical and practical significance.
Theoretically, this result proves the important value of discipline-specific learning behavior indicators in academic performance prediction. Most existing academic performance prediction studies rely on general learning behavior indicators and ignore the learning characteristics and behavioral differences of different disciplines. The results of this study show that general learning behavior indicators can only explain part of the variation in academic performance, while discipline-specific learning behavior indicators can explain additional variation that cannot be explained by traditional general indicators. This study found that music-specific cyclic listening behavior indicators can explain an additional 16.4% of the variation in academic performance, and this proportion even reaches 28.5% in piano majors. This finding provides an important reference for academic performance prediction research in other disciplines. Future research should pay more attention to discipline-specific learning behavior indicators to further improve the accuracy of academic performance prediction.
Practically, this result provides a new approach for personalized learning intervention in the field of music education. By analyzing students’ cyclic listening behavior data, teachers can timely identify students with learning difficulties and carry out targeted intervention and guidance. For example, for students with low deep cyclic rates, teachers can guide them to conduct more in-depth music listening, provide them with detailed listening guidance and question lists, and help them focus on key elements such as the structure, harmony, and emotion of musical works during the listening process. For students with high fragmented cyclic rates, teachers can help them develop good learning habits, improve their learning concentration, and guide them to conduct complete and systematic music listening.
In addition, the high-precision prediction model constructed in this study can be integrated into online music learning platforms to provide students with personalized learning resource recommendations and learning path planning. For example, the platform can recommend musical works suitable for their level and interests based on students’ cyclic listening behavior data, or remind them to conduct more in-depth listening to certain important works. The platform can also formulate personalized learning plans for students according to their listening habits, reasonably arrange listening time and content, and improve learning efficiency.
The results of this study also have important implications for music education assessment. Traditional music education assessment mainly relies on final examination scores, which is lagging and one-sided and cannot fully reflect students’ learning processes and learning efforts. Incorporating cyclic listening behavior data into the music education assessment system can realize the organic combination of formative assessment and summative assessment, and evaluate students’ learning effects more comprehensively and objectively. For example, teachers can take indicators such as students’ deep cyclic rate and total duration of cyclic listening as part of their usual grades to encourage students to actively participate in music listening activities and improve their learning initiative and enthusiasm.
- Conclusion
7.1 Main Research Findings
This study systematically explores the improvement effect of music-specific cyclic listening behaviors on the prediction accuracy of music academic performance for the first time. Through empirical analysis of multi-dimensional learning behavior data and academic performance data from 314 music majors, the following three core conclusions are drawn:
First, music-specific cyclic listening behavior is a discipline-specific behavior with unique value in music learning. This study constructs a scientific quantitative indicator system from four dimensions: behavior frequency, duration, depth, and pattern, and clearly distinguishes two distinct listening modes: “effective deep cyclic listening” and “ineffective fragmented cyclic listening”. The study found that there are significant differences in cyclic listening behaviors among students of different majors. The deep cyclic rates of piano and vocal music majors are significantly higher than those of instrumental music and music theory majors, which is highly consistent with the learning characteristics and ability requirements of each major.
Second, music-specific cyclic listening behaviors have a significant independent predictive effect on music academic performance. After controlling for traditional general learning behavior indicators such as class attendance rate, homework completion rate, and instrument practice duration, music-specific cyclic listening behaviors can still explain 16.4% of the variation in academic performance. Among them, the deep cyclic rate is the most important factor affecting music academic performance (β=0.312, p<0.001), and its predictive power even exceeds that of instrument practice duration; while the fragmented cyclic rate has a significant negative impact on music academic performance (β=-0.098, p<0.05). This result indicates that purposeful and systematic in-depth listening is a key success factor in music learning, while fragmented superficial listening is not conducive to the development of musical ability.
Third, incorporating music-specific cyclic listening behavior indicators can significantly improve the accuracy of music academic performance prediction models. For all four machine learning algorithms, the performance of the fusion feature set model is significantly better than that of the traditional general feature set model. Among them, the R² of the best-performing Gradient Boosting Regression model increased from 0.534 to 0.634, an increase of 18.7%; MAE decreased from 4.765 to 3.892, a decrease of 15.3%. This improvement varies significantly among different majors. The R² improvement rate is the largest for piano majors (28.5%), followed by vocal music majors (25.3%), and the smallest for music theory majors (9.9%). This result fully proves the important value of discipline-specific learning behavior indicators in academic performance prediction.
7.2 Theoretical Contributions
The theoretical contributions of this study are mainly reflected in the following three aspects:
First, it expands the research boundary of music learning behavior theory. This study systematically defines the concept of “music-specific cyclic listening behaviors” for the first time, and constructs a quantitative indicator system including four dimensions: frequency, duration, depth, and pattern, making up for the deficiency of existing studies in the quantification of music listening behaviors. At the same time, this study reveals the differential impacts of different cyclic listening modes on academic performance, clarifies the core position of “deep cyclic listening” in music learning, and provides a new theoretical perspective for understanding the music learning process.
Second, it enriches the application of the theory of learning behavioral engagement in the field of music education. This study incorporates music-specific cyclic listening behaviors into the analytical framework of learning behavioral engagement, and proves that discipline-specific behaviors are an important part of learning behavioral engagement. The study found that cyclic listening behavior, as an active learning behavior, can better reflect students’ learning engagement than passive behaviors such as class attendance. This finding improves the application of the theory of learning behavioral engagement in specific disciplinary fields.
Third, it provides a new research paradigm for academic performance prediction theory. This study breaks through the limitation of existing studies that mainly rely on general learning behavior indicators, and proves that discipline-specific learning behavior indicators can significantly improve the accuracy of academic performance prediction. This finding provides an important reference for academic performance prediction research in other disciplines, and promotes the transformation of academic performance prediction research from a “general paradigm” to a “discipline-specific paradigm”.
7.3 Practical Implications
The results of this study have important guiding significance for music education practice:
For music educators, they should fully attach importance to the important role of cyclic listening behaviors in music learning and incorporate listening guidance into daily teaching content. Teachers can formulate clear listening tasks and requirements for students to guide them to conduct purposeful and in-depth music listening; at the same time, by analyzing students’ cyclic listening behavior data, they can timely identify students with learning difficulties and carry out targeted intervention and guidance. For example, for students with low deep cyclic rates, detailed listening guides and question lists can be provided; for students with high fragmented cyclic rates, they can be helped to develop good learning habits and improve their learning concentration.
For developers of online music learning platforms, they should make full use of the massive listening behavior data generated by digital platforms to develop data-driven personalized learning functions. The platform can recommend musical works suitable for their level and interests based on students’ cyclic listening behavior data; provide students with personalized learning plans and progress reminders; and visually display their own listening behavior data to students to help them understand their learning status and adjust their learning strategies.
For education managers, they should reform the traditional music education assessment system and organically combine process assessment with summative assessment. Students’ cyclic listening behavior data can be included in the assessment scope of usual grades to encourage students to actively participate in music listening activities and improve their learning initiative and enthusiasm. At the same time, the academic performance prediction model is used to dynamically monitor students’ learning performance, identify learning risks in advance, and take preventive intervention measures to improve the overall quality of education.
7.4 Research Limitations and Prospects
Although this study has obtained some valuable findings, it still has certain limitations. First, the sample of this study only comes from music undergraduates in 3 institutions of higher education, and the representativeness of the sample is limited. Future research can expand the sample range to include students of different levels and age groups to verify the universality of the research results. Second, this study adopts a cross-sectional research design, which can only reveal the correlation between variables. Future research can adopt a longitudinal research design to track the dynamic changes of students’ cyclic listening behaviors and academic performance, and further explore the causal relationship between them. Finally, this study does not consider the influence of factors such as the type, difficulty, and style of musical works. Future research can further explore how these factors moderate the relationship between cyclic listening behaviors and academic performance.
Future research can also be carried out from the following aspects: First, further improve the quantitative indicator system of music-specific cyclic listening behaviors, and add more indicators reflecting listening quality, such as the degree of attention concentration during listening and the depth of understanding of musical works. Second, adopt a multimodal data fusion method to combine listening behavior data with physiological data, eye movement data, etc., to more comprehensively understand students’ music learning processes. Third, develop a personalized learning intervention system based on cyclic listening behavior data, and verify its effectiveness through experimental research. Fourth, extend the research to other art disciplines such as fine arts and dance, explore the role of discipline-specific learning behaviors in academic performance prediction, and construct a more universal theoretical framework for discipline-specific academic performance prediction.
References
[1] Luo C. The effect of listening to music while studying on cognitive performance: A meta-analysis[J]. Advances in Psychological Science, 2024, 32(5): 892-903.
[2] Wang Y, Li X, Zhang H. Listening to music while studying: Background sound and its relationship to task performance[J]. Music & Science, 2022, 5: 1-12.
[3] Smith A, Jones B. Does listening to music help or hurt studying? A systematic review[J]. Educational Psychology Review, 2026, 38(1): 45-67.
[4] Wang Y L. The effect of different playback modes on college students’ attention stability and memory effect[J]. Chinese Journal of Health Psychology, 2023, 31(8): 1245-1250.
[5] Chen Y Q. The effect of background music on college students’ vocabulary memory[J]. Psychological Science, 2015, 38(3): 623-628.
[6] Hassan Y A. The harmonious balance: Exploring the impact of music on concentration and academic performance[J]. Journal of Educational Research, 2026, 119(2): 123-135.
[7] Mahmood S, Ahmed M, Ali S. Sustainable engagement and academic achievement under impact of academic self-efficacy through mediation of learning agility—Evidence from music education students[J]. Frontiers in Psychology, 2025, 16: 923746.
[8] Zhang X, Li M. Theoretical construction of learning engagement mechanism: Connotation evolution, influencing factors and enhancement mechanism[J]. Educational Research, 2025, 46(11): 78-89.
[9] Fredricks J A, Blumenfeld P C, Paris A H. School engagement: Potential of the concept, state of the evidence[J]. Review of Educational Research, 2004, 74(1): 59-109.
[10] Gabriele W, Prinz W. Directing attention to movement effects enhances motor learning[J]. Journal of Motor Behavior, 2001, 33(2): 169-179.
[11] Levitin D J. This is your brain on music: The science of a human obsession[M]. New York: Dutton, 2006.
[12] Bjork R A, Bjork E L. A new theory of disuse and an old theory of stimulus fluctuation[M]. Hillsdale, NJ: Erlbaum, 1992.
[13] Liu C. From practicing to mastering: Unlocking the scientific method of piano repetitive practice[J]. Piano Artistry, 2026, (1): 34-39.
[14] Margulis E H. On repeat: How music plays the mind[M]. Oxford: Oxford University Press, 2014
[15] Kanouse B. Play it again: People find comfort listening to the same song over and over[J]. Neuroscience News, 2018, 2(16): 1-5.
[16] Zhan Z M, Ao Z J, Li X, et al. A comparative study of student performance prediction models based on machine learning[J]. Advances in Education, 2026, 16(1): 1159-1169.
[17] Li Y, Wang Z, Zhang L. The construction of student-centered artificial intelligence online music learning platform based on deep learning[J]. Scientific Reports, 2025, 15(1): 95729.
[18] Wang H, Li N. Research on personalized prediction of high school students’ music learning effect based on deep learning model[J]. China Educational Technology, 2025, (10): 112-119.
[19] Zhang H, Liu Y, Chen X. Based on machine learning of music teaching effect prediction model construction and application[J]. Journal of Intelligent & Fuzzy Systems, 2025, 49(6): 12345-12356.
[20] Chen S, Li J, Wang Y. Big data analysis of music education for future intelligent learning leveraging e-learning using deep learning technology[J]. Computer-Aided Design and Applications, 2024, 21(S2): 103-114.