PREDICTING RIDE-HAILING CUSTOMER LOYALTY USING RANDOM FOREST AND PRINCIPAL COMPONENT ANALYSIS

Abstract

Customer loyalty is important for ride-hailing companies because users can easily change from one platform to another. This study applied Random Forest to classify the loyalty of ride-hailing users in Pekanbaru, Indonesia, and examined whether Principal Component Analysis (PCA) improved the result. The original dataset contained 200 questionnaire records covering demographic characteristics, trust, innovation, service quality, customer satisfaction, and customer loyalty. After seven duplicate records were removed, 193 records were divided into 154 training records and 39 test records. The baseline model achieved 0.67 accuracy and a weighted F1-score of 0.67. The best PCA model retained 80% of the variance, used seven trees, and achieved 0.69 accuracy with a weighted F1-score of 0.68. Service quality, customer satisfaction, trust, and innovation were the most important predictors. The model performed best for the Loyal class, while the smaller classes were harder to identify. One loyalty category was absent from the test set; therefore, the findings should be treated as an initial result rather than a model ready for operational use.