Prediction of Student Academic Stress Levels Using the Decision Tree Algorithm and Particle Swarm Optimization

Abstract

Academic stress was recognized as a major challenge for university students because it negatively affected learning outcomes, mental health, and overall well-being. The purpose of this research was to develop and validate a predictive model of student academic stress levels and to evaluate whether optimization techniques improved the performance of a baseline classifier. Data were collected from 413 active students of Universitas Sapta Mandiri from the 2022 and 2023 cohorts using the Perception of Academic Stress (PAS) scale, which consisted of 18 indicators, together with demographic, academic, and psychosocial attributes. The Decision Tree (DT) algorithm was selected for its interpretability and transparency in multi-class classification. To improve generalization, its parameters were optimized using Particle Swarm Optimization (PSO) with 10 particles and 20 iterations. The baseline model achieved an accuracy of 93 percent, with the highest recall observed in the low-stress group. After optimization, the accuracy increased to 95 percent, and the recall for the high-stress group reached 0.96, indicating greater sensitivity to students at risk. These results confirmed that the research objectives were achieved, as the integration of DT with PSO enhanced both accuracy and class balance. The proposed model was consistent with the intended purpose of supporting early detection and timely academic and psychological interventions in higher education institutions.