Deep Learning and Traditional Algorithms: A Comparative Study on Predicting Second-Hand Car Prices

As the used car market is rapidly increasing, the ability to accurately forecast prices have become increasingly important to both buyers and sellers. Using a variety of predictive models, including Random Forest, XGBoost, LightGBM, K-Nearest Neighbors (KNN), Lasso Regression (LR), Ridge Regression (RR) and Neural Networks (NN), the study seeks to develop machine learning models that are accurate and reliable in determining the price of a used car. We collect data from CarsWale for the web scraping process. We are able to analyze the details of the vehicle set, such as kilometers driven, year of production, fuel type, engine capacity, and ownership details. To begin with, our methodology included data pre-processing and feature engineering, and then we implemented regression and ensemble learning models. Random Forest and Neural Networks are experiments with mathematical models, and they show forecast accuracy, while KNN is managed quite well with the noisy and multidimensional data. The amalgamation of some artificial neural networks, especially one that relies on XGBoost and LightGBM, resulted in the prediction of the prices with acceptable accuracy. The models, such as linear regression and decision trees, outperformed a hybrid framework of XGBoost and LightGBM with a peak R2 of 96.52% and error being the lowest. The results point to the advantages of coupling the different machine learning solutions to encounter the vagueness of the second-hand car pricing that is affected by factors such as the condition of the vehicle, its mileage, and market fluctuations. The paper allows individuals to explore the feasibility of their investment mostly in the area of automobiles by focusing on the used car scenario. The paper attempts to demonstrate by modeling that the models are not over-specialized for specific markets and that they are flexible and scalable in various economic setups.