Support Vector Machine Hyperparameter Tuning for Student Addiction Prediction

Abstract

Student addiction is a severe condition with far reaching academic, psychological, and societal implications. Early identification of students at risk for substance use disorders (SUD) enables timely intervention and may reduce long-term impacts. While machine learning has been applied to predict addiction behaviors, few studies have systematically examined the impact of Support Vector Machine (SVM) kernel selection, regularization parameter tuning (C), and polynomial degree adjustment. In this work, we applied SVM classifiers to predict student addiction using behavioral, academic, and social attributes. For computational efficiency, the experiments were conducted on the first 2000 rows of the dataset, as extending to the full dataset did not yield significant differences in performance. We systematically tested different kernel functions (linear, radial basis function, polynomial), values of C, and polynomial degrees, and evaluated their effect on model accuracy. Among all combinations, several configurations including the linear kernel, RBF kernel with C=1, and polynomial kernels with degrees 2–3 achieved the highest accuracy of 78.9%.To increase the biological relevance of our study, we incorporate a neuroscience-based overview of addiction, highlighting how repeated drug use alters brain chemistry, particularly dopamine and opioid pathways, leading to impaired impulse control and increased risk-taking behavior. By integrating biological context with machine learning optimization, this study contributes a novel perspective to student health analytics and emphasizes the importance of model calibration in improving early risk prediction within education and public health domains.