Scholarlink

Accurate price forecasting for used vehicles has become crucial in the $1.4 trillion global used car market, addressing information asymmetry between buyers and sellers. This study evaluates machine learning techniques for old car price prediction, comparing Random Forest, XGBoost, and hybrid models using a dataset of 8,740 vehicles. Key findings reveal Random Forest achieved superior performance with 0.28 RMSE and 0.92 R2 score, significantly outperforming linear regression (2.14 RMSE). Critical predictive features included mileage (23.7% importance), vehicle age (18.9%), and engine power (15.4%). The XGBoost model demonstrated 97.53% accuracy in optimized implementations, particularly effective for luxury vehicles where traditional methods show 23% higher errors. These results align with recent findings from the Journal of Decision Analytics and Intelligent Computing [1], confirming tree-based ensembles’ superiority in handling non-linear automotive valuation patterns. The study provides actionable insights for developing dynamic pricing tools and informs policy-making for fair market practices.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Forecasting Old Car Prices Through Machine Learning Techniques

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sonali Dadwal,
  • Lakshmojee Koduru,
  • Riya Mondal,
  • Shanu Khare,
  • Payal Thakur,
  • Navjot Singh Talwandi

摘要

Accurate price forecasting for used vehicles has become crucial in the $1.4 trillion global used car market, addressing information asymmetry between buyers and sellers. This study evaluates machine learning techniques for old car price prediction, comparing Random Forest, XGBoost, and hybrid models using a dataset of 8,740 vehicles. Key findings reveal Random Forest achieved superior performance with 0.28 RMSE and 0.92 R2 score, significantly outperforming linear regression (2.14 RMSE). Critical predictive features included mileage (23.7% importance), vehicle age (18.9%), and engine power (15.4%). The XGBoost model demonstrated 97.53% accuracy in optimized implementations, particularly effective for luxury vehicles where traditional methods show 23% higher errors. These results align with recent findings from the Journal of Decision Analytics and Intelligent Computing [1], confirming tree-based ensembles’ superiority in handling non-linear automotive valuation patterns. The study provides actionable insights for developing dynamic pricing tools and informs policy-making for fair market practices.