A multi-factor machine learning framework for predicting and profiling student academic performance using behavioral,...

This study introduces a multi-factor machine learning framework designed to predict and profile student academic performance using behavioral, financial, and wearable data. The dataset, collected from higher education students in India, integrates lifestyle behaviors (e.g., activity score, screen time), financial variables (income, scholarship, loans, and tuition), and physiological data such as average heart rate, high BPM, and smartwatch usage. The method begins with structured preprocessing and feature engineering, including the construction of a Financial Stress metric and a composite stress index combining financial and physiological inputs. Multiple regression models were benchmarked to predict CGPA, with Random Forest yielding the highest accuracy (R² ≈ 0.30). Further, wearable-related features were analyzed using correlation and t-tests to examine their relationship with academic outcomes. Finally, unsupervised clustering techniques (K-Means and Agglomerative Clustering) were employed to segment students into interpretable academic and stress-risk profiles. Model accuracy and cluster quality were validated using R², RMSE, MAE, silhouette scores, and PCA-based visualizations. This method supports early detection of at-risk students and provides an adaptable blueprint for educational institutions to implement predictive academic analytics. Predicts CGPA using lifestyle, financial, and wearable metrics Profiles students into risk-based academic clusters Enables interpretable and reusable machine learning pipeline for education.