Machine Learning-Based Student Performance Prediction System: Comparative Analysis and Practice of Six Regression Models

This article introduces a project that uses multiple machine learning regression algorithms to predict student academic performance. By comparing the performance of six models including linear regression, random forest, and decision tree, it is finally determined that multiple linear regression with an R² score of 98.84% is the optimal solution. The project aims to provide educational institutions with data-driven student performance prediction tools to help identify students in need of support and optimize teaching strategies.

In the field of higher education, accurately predicting student academic performance is a focus of educators and data scientists. Traditional evaluation methods rely on single indicators or subjective judgments, making it difficult to comprehensively capture multi-dimensional influencing factors. The team from Sri Lanka Institute of Information Technology (SLIIT) developed this project with the goal of analyzing key factors affecting student performance, comparing the predictive capabilities of multiple regression algorithms, selecting a model suitable for deployment, and providing data support for educational institutions.

The project uses a cleaned dataset containing multi-dimensional features:

• Study Duration: Weekly study hours
• Past Grades: Previous exam scores
• Extracurricular Activities: Participation status
• Sleep Duration: Daily sleep hours
• Mock Exam Practice Volume: Number of completed mock exams
• Learning Efficiency: Comprehensive indicator The target variable is the Performance Index, which quantifies the student's overall academic level. Feature design reflects the principle of multiple factors acting together.

The project evaluated six mainstream regression algorithms:

1. Multiple Linear Regression (MLR): A classic method with strong interpretability and efficient computation
2. Random Forest Regression: Ensemble learning that captures non-linear interactions and is insensitive to outliers
3. Decision Tree Regression: Intuitive and easy to understand, but prone to overfitting
4. Gradient Boosting Regression: Boosting strategy with good performance on structured data, but long training time
5. Support Vector Regression (SVR): Stable for small samples, but inefficient for large-scale data
6. Multi-Layer Perceptron (MLP): Neural network with strong expressive power, but requires parameter tuning to avoid overfitting

Three metrics (MAE, RMSE, R²) were used for evaluation:

Model	MAE	RMSE	R²
Multiple Linear Regression	1.6466	2.0753	0.9884
MLP Regressor	1.6707	2.1024	0.9881
Support Vector Regression	1.6805	2.1220	0.9879
Gradient Boosting Regression	1.7034	2.1418	0.9877
Random Forest Regression	1.9511	2.4345	0.9841
Decision Tree Regression	2.0435	2.5755	0.9822
The results show that multiple linear regression is optimal with an R² score of 0.9884, explaining approximately 98.84% of the variance.

The project uses a Python toolstack:

• Data Processing: Pandas, NumPy
• Machine Learning: Scikit-learn
• Visualization: Matplotlib, Seaborn
• Model Persistence: Joblib
• Development Environment: Jupyter Notebook The project structure is clear, including data directories, notebooks, visualization charts, and technical reports, reflecting good engineering practices.

The project provides important insights:

• Model Selection: Prioritize simple models; linear models perform better when features are linearly correlated
• Interpretability: In educational scenarios, interpretability is more important than precision; linear model coefficients reflect the influence of factors
• Feature Engineering: Data and features determine the upper limit, and models approach this limit The application value lies in helping educational institutions identify students in need of support and optimize teaching strategies.

Future improvement directions:

• Expand feature dimensions (socioeconomic background, mental health, etc.)
• Explore integration strategies (model stacking)
• Develop a web interface
• Compare with deep learning Conclusion: The project demonstrates the application of machine learning in the education field, verifies the wisdom of prioritizing simple models, and multiple linear regression, which balances performance, efficiency, and interpretability, is an ideal deployment choice, providing a complete data science case for learners.