A Novel Concept Study on Key Factors Affecting Student Performance Using Machine Learning Algorithms

Abstract

This study examines the correlation between critical aspects, including study hours, sleep 
hours, attendance rate, dietary habits, and extracurricular contribution, and their impact on 
assessment performance through various machine learning algorithms. Five models were examined 
to forecast student performance: Linear Regression (LR), Decision Tree (DT), Random Forest (RF), 
Support Vector Machine (SVM), and XGBoost.  Data was generated for over 300 students, with 
each feature exerting a direct or indirect impact on the assessment performance. Each model has 
been developed on a training dataset and tested on a distinct test set to measure performance. The 
evaluation employed two principal metrics: Root Mean Squared Error (RMSE) and R-squared (R²). 
The XGBoost model exhibited optimal performance, evidenced by the lowest RMSE and maximum 
R², signifying its capacity to explain intricate relationships within the data. The Random Forest and 
Decision Tree models demonstrated favorable outcomes, with the Random Forest exhibiting greater 
resilience to overfitting compared to the Decision Tree. The SVM model, while successful, exhibited 
reduced efficiency owing to the non-linear correlations present in the data. Although interpretable, 
the Linear Regression model exhibited fewer errors than the alternative models. The study also 
examined learning curves, feature significance, and residual analysis to better understand the 
model's strengths and limitations. Scatterplots and histograms were employed to evaluate model 
performance and discern important features. The results indicate that XGBoost is a very efficient 
tool for performance forecasting, particularly in educational datasets characterized by numerous 
factors. This study evaluates machine learning models and illustrates how real-world variables 
influencing student performance can potentially be efficiently analyzed using complex algorithms, 
offering useful insights for educational stakeholders seeking to enhance student outcomes