Analisis Prediksi Durasi Pengiriman Barang Menggunakan Algoritma Random Forest pada PT Herona Express Jember

Abstract

On-time delivery is one of the most critical logistics performance indicators and a key determinant of customer satisfaction. However, PT Herona Express Jember has so far estimated delivery duration manually based on operator experience, making it inconsistent and unable to adapt to variations in route, transit conditions, and delivery volume fluctuations. This study aims to analyze the characteristics of historical delivery data, evaluate the performance of the Random Forest Regression algorithm in predicting delivery duration, and identify the features that most influence that duration. The data were obtained from the Herona Information System (HIS) for the 2025 period, comprising 4,743 rows, and underwent preprocessing that included data cleaning, feature transformation, and feature selection, leaving 4,174 records with seven selected features: origin city, destination city, distance, number of packages, service type, transit status, and shipping day. The data were split into 80% training and 20% testing, and the model was optimized using GridSearchCV with 5-fold cross-validation. Evaluation used the MAE, RMSE, and R-squared metrics. The testing results show that the model produced an MAE of 0.4036 days, an RMSE of 0.5501 days, and an R-squared of 0.3942, meaning that the average prediction error is below half a day. The feature importance analysis identified transit status as the most influential feature (26.80%), followed by shipping day (24.32%) and distance (15.12%). This study concludes that the Random Forest algorithm is able to predict delivery duration with a low error rate, and that the presence or absence of a transit process is the operational factor that most determines the length of delivery duration, so that the model results can serve as an initial reference for estimating delivery duration more objectively.