การพยากรณ์ราคารถยนต์มือสองโดยใช้เทคนิคการสกัดข้อมูลจากเว็บไซต์และแบบจำลองการเรียนรู้ของเครื่อง
Year: 2026
Type: article
Abstract: This study aimed to develop a predictive model for used car prices in Thailand using web scraping techniques and machine learning algorithms. Data were collected from Kaidee.com, One2Car.com, and Chobrod.com, totaling 55,989 records. After data cleaning, 42,823 valid records remained, containing 13 fundamental attributes for model construction. Two experiments were conducted: (1) using only the basic features and (2) incorporating six newly engineered features—car age, annual usage rate, squared mileage, squared engine size, cumulative usage load, and temporal load—to enhance the model’s learning capability. The performance of five models, including XGBoost, Random Forest, LightGBM, CatBoost, and Gradient Boosting, was compared using MAE, RMSE, MAPE, R², and accuracy. The results showed that XGBoost achieved the best prediction performance. With the additional features, the R² value improved from 0.9262 to 0.9419, accuracy increased from 89.42% to 92.38%, and MAPE decreased from 10.58% to 7.62%, indicating that feature engineering significantly enhanced model accuracy. Feature importance analysis revealed that the most influential factors affecting used car prices were fuel type, car type, engine size, brand, and squared engine size. The findings confirm that integrating machine learning with feature engineering substantially improves predictive performance and can serve as a decision-support tool for buyers, sellers, and financial institutions to promote transparency and fairness in Thailand’s used car market.
Institution: Phuket Rajabhat University
Language: Thai
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7215419342