COMPARATIVE ANALYSIS OF ENSEMBLE CLASSIFICATION MODELS AND SUPPORT VECTOR MACHINES IN MEASURING STRESS LEVELS BASED ON EEG SIGNALS

Seftia Angelina (Universitas Prima Indonesia)
Sau Dohot Siregar (Universitas Prima Indonesia)
Achmad Ridwan (Universitas Prima Indonesia)
Lewis Francolim (Universitas Prima Indonesia)

Jiko (Jurnal Informatika dan komputer)
Vol 9 No 1 (2026)

Stress is a physiological and psychological response that can develop into serious health issues when prolonged. EEG-based stress detection has become an important approach; however, many studies still lack validation for multilevel classification and real-world conditions. This study focuses on inmates at Binjai Correctional Facility and compares the performance of Support Vector Machine (SVM), Random Forest (RF), and a combined ensemble model of Random Forest and AdaBoost for classifying three stress levels: stressed, relaxed, and neutral, using EEG signals. Experimental results show that the SVM model achieved an accuracy of 81% with a Minimum Classification Error (MCE) of 0.16. The Random Forest model significantly improved performance, reaching 96% accuracy and an MCE of 0.04. The best performance was obtained by the ensemble model combining Random Forest and AdaBoost, which achieved an accuracy of 97% and reduced the MCE to 0.03, indicating a 1% improvement over Random Forest alone.