ROAD DAMAGE PREDICTION USING NAÏVE BAYES AND RANDOM FOREST: A CASE STUDY OF PPK 1.5 DIRECTORATE GENERAL OF HIGHWAYS, SOUTH SUMATRA PROVINCE

Good road conditions are essential for smooth transportation, community mobility, and economic activity, yet historical road handling data managed by the Commitment Making Officer (PPK) 1.5 of the Directorate General of Highways in South Sumatra Province is still managed manually, causing road handling to be reactive, difficult to monitor centrally, and prone to inaccurate budget estimation. This study aims to design and build a web-based application predicting road damage levels using Naive Bayes and Random Forest, and predicting maintenance work volume using Random Forest Regressor. The study used the Rapid Application Development (RAD) method, consisting of Requirements Planning, User Design, Construction, and Cutover stages, with 5,138 historical maintenance records from 2022 to 2025 on the Betung–Palembang City Boundary road section, designed using a Flowchart, Data Flow Diagram (DFD), and Entity Relationship Diagram (ERD). The application was successfully built with core features for survey input, PPK verification, dataset management, damage prediction, and reporting. Black Box Testing confirmed all functional modules operated as expected, and User Acceptance Testing involving seven stakeholders produced an average score of 96.27% (very good category). The ensemble Naive Bayes–Random Forest model achieved 95% Accuracy, Precision, Recall, and F1-Score, higher than standalone Naive Bayes (91%), while the Random Forest Regressor produced R² values between 0.90 and 0.93 across four maintenance work types. These findings indicate that the SIKERJA application is feasible and has the potential to support digitalization of the road damage prediction process and improve the effectiveness of road maintenance decision-making at PPK 1.5.