Behavioral, Lifestyle, and Mental Health Factors Associated With Academic Performance: A Cognitive-Behavioral and Multi-Dataset Machine Learning Analysis
DOI:
https://doi.org/10.67440/ahj.vi.2186Keywords:
Cognitive Behavioral Therapy; Academic Performance Prediction; Machine Learning in Education; Student Mental Health; Feature Importance; Educational Data Mining; Random Forest; Multi-Dataset Analysis.Abstract
Background: Identifying which behavioral, lifestyle, and mental health factors most reliably predict adolescent and young adult academic performance remains a critical challenge for evidence-based educational intervention. Cognitive Behavioral Therapy (CBT) provides a structured framework for understanding how cognition, behavior, and emotion jointly shape learning outcomes. Yet systematic, data-driven identification of CBT-relevant predictors across heterogeneous student populations is lacking.
Methods: Four publicly available Kaggle datasets (total n = 13,564; DS1: student habits, n = 1,000; DS2: lifestyle patterns, n = 2,000; DS3: university grade factors, n = 10,064; DS4: mental health, n = 500) were analyzed. Thirty-five features were mapped to four CBT domains: behavioral engagement, emotional regulation, physiological factors, and environmental support. Five machine learning (ML) algorithms—Linear Regression, Logistic Regression, Random Forest (RF), Support Vector Machine (SVM), and Gradient Boosting—were benchmarked on regression and classification tasks. Feature importance was extracted per dataset and synthesized across the combined corpus (n = 13,564) using normalized impurity-based and coefficient-based metrics.
Results: RF achieved near-perfect mental health risk classification (DS4: accuracy = 98.00%, F1 = 0.978, AUC = 1.000; CV F1 = 0.974 ± 0.011). Linear Regression best explained continuous exam scores (DS1: R² = 0.8968, RMSE = 5.15). Across the combined dataset, Study Hours (importance = 0.1581), Attendance (0.1558), Screen Time (0.1527), Sleep Hours (0.1485), and Stress Level (0.1101) emerged as the five dominant predictors. Depression score (0.4222) and Anxiety score (0.2416) dominated the mental health classification task exclusively.
Conclusion: Behavioral engagement (study hours, attendance) and screen time management consistently outrank psychological constructs as cross-dataset predictors of academic performance, providing a prioritized, empirically grounded target hierarchy for CBT-informed educational interventions.
