<![CDATA[Urban traffic congestion in Indonesian cities is closely linked to student commuting, yet the behavioral factors underlying mode choice are numerous and strongly correlated, which limits conventional classification models. This study integrates Principal Component Analysis (PCA) with a Random Forest (RF) classifier to predict transportation mode choice among university students in Medan, Indonesia. Data were obtained from 1,177 valid questionnaire responses covering 39 items, which were aggregated into eight behavioral constructs; seven were retained after sampling-adequacy screening. Applying the Kaiser criterion, two principal components were extracted, jointly explaining 70.42% of the total variance, and used as predictors of six transportation modes. The RF model attained 66.09% accuracy and a Cohen's Kappa of 0.5714, indicating moderate agreement between predicted and observed choices. Performance differed markedly across modes, from an F1-score of 82.37% for Mini Bus to 17.65% for Online Car Transportation, which was frequently misclassified as Online Motorcycle Transportation. The findings indicate that a compact two-component representation retains substantial behavioral information while yielding moderate predictive performance, and that mode-specific data enrichment is needed before such models can inform campus transportation planning.]]>
Copyrights © 2026
