A Comparative Analysis of Tree-Based Models and Deep Learning for Predicting Used Car Prices: A Standardized Study Utilizing Indonesian Marketplace Data

The Indonesian used-car market is expanding rapidly, yet price transparency remains elusive due to absent standardized valuation frameworks and seller-driven information asymmetry in peer-to-peer marketplaces. This study proposes a machine learning framework for automated used car price prediction, addressing two gaps in the Indonesian literature: inconsistent preprocessing pipelines and the absence of comprehensive multi-algorithm evaluation — including deep learning — on clean local data. A dataset of 29,747 listings was scraped from OLX and Mobil123 across JABODETABEK, covering eight brands (Honda, Toyota, Daihatsu, Suzuki, Mercedes-Benz, Wuling, BYD, Mitsubishi) and cleaned to 16,291 records including 410 battery-electric vehicles. Three contributions are presented: (1) a reproducible preprocessing pipeline with evidence-gated model/variant standardization, enabling fair cross-model comparison; (2) the first six-algorithm evaluation — including MLP — on clean Indonesian used car data, with XGBoost achieving R² = 0.9647, MAPE = 6.48%, MAE = Rp 15,152,085, empirically confirming that tree-based models outperform deep learning on structured tabular data; and (3) a web-based deployment with confidence ranges and 5-year depreciation projections. Two temporal features — Vehicle Age and Time Index — are evaluated over a 20-month collection window, enabling XGBoost to more effectively capture depreciation and market-trend dynamics. SHAP-based interpretability complements the analysis.