PREDIKSI KESEHATAN MENTAL REMAJA BERDASARKAN KEBIASAAN DIGITAL MENGGUNAKAN METODE LOGISTIC REGRESSION
Abstract
Adolescent mental health is an important issue in the digital era because device usage duration, social
media intensity, and digital fatigue may influence psychological conditions and social interaction. This
study aims to analyze digital habit patterns, identify the association between digital habit indicators and
mental health risk, and develop a Logistic Regression prediction model among adolescents at HKBP
Bethesda Ressort Bethesda. Data were collected using an 18-item questionnaire from 200 respondents
through total sampling. The analysis included data-structure detection, value cleaning, ordinal encoding,
calculation of Antisocial, Anxiety, and Individual scores, binary-label formation, feature standardization,
stratified 80:20 splitting, model training with balanced class weights, and evaluation using a confusion
matrix and classification metrics. The mean scores for Antisocial, Anxiety, and Individual were 9.82,
7.05, and 6.63, respectively. On 40 test observations, the model achieved 87.50% accuracy, 100% recall
for the at-risk class, an F1-score of 0.44, and an AUC-ROC of 0.95. The Odds Ratio analysis indicated
the strongest contribution from Individual at 8.00, followed by Antisocial at 5.36 and Anxiety at 4.83.
The model may support initial screening, but its results must be interpreted cautiously because only 11
respondents were labeled at risk.

