Conceptualizing Student Risk: A Dual-Objective Machine Learning Model and System Architecture for Academic and Social Media Outcomes

Abstract

Social media has become deeply embedded in students’ daily lives, influencing how they manage time, emotions, and academic responsibilities. However, most research continues to examine academic performance, digital habits, and psychological well-being in isolation. This study introduces a dual-objective machine learning framework designed to jointly predict two outcomes: (i) academic success based on pre-admission records and (ii) negative social media impact using behavioural and mental-health indicators. The data were drawn from a higher education institution. Pre-admission records included demographic details, interview scores, and subject-specific marks, while a structured questionnaire captured perceptions of social media interference with sleep, sleep quality, mood, and stress or anxiety. Binary outcomes were defined for academic success (Successful / Not Successful) and social media impact (Negative Impact / No Impact) using institutional thresholds. After preprocessing with label encoding and median imputation, five classifiers—Random Forest, Support Vector Machine, multi-layer perceptron neural network, Gradient Boosting, and a soft-voting ensemble—were trained with an 80/20 split.Gradient Boosting consistently outperformed others, achieving 100% accuracy for both tasks. Random Forest followed closely, while Support Vector Machine and neural networks struggled with minority classes. Results confirm the effectiveness of gradient boosting and highlight the value of integrating behavioural and mental-health features for AI-driven student risk analytics.