<![CDATA[This study aims to analyze the capability of an Artificial Neural Network (ANN) model in detecting stress levels based on digital behavioral data. The dataset comprises 3,685 Indonesian respondents represented through six digital behavior variables (X1–X6) covering smartphone and social media usage patterns, and six psychological stress symptom variables (S1–S6) based on an adapted PSS-10. A Multilayer Perceptron (MLP) was trained using a two-stage experimental design: Stage 1 using only X1–X6, Stage 2 integrating X1–X6 and S1–S6. Evaluation involved 27 parameter combinations per stage. Results show Stage 1 failed with mean accuracy 77.34% (Macro F1=0.29) due to insufficient discriminative signal, while Stage 2 succeeded with peak accuracy 98.10% and Macro F1=0.97 at optimal configuration (80:20 split, k=8, hidden=16). The +18.16% improvement quantifies the mediation role of psychological variables in the transactional stress model.]]>
Copyrights © 2026
