Car Price Prediction using Machine Learning Techniques

Gegic, Enis; Isakovic, Becir; Keco, Dino; Masetic, Zerina; Kevric, Jasmin · 2019 · OpenAlex-citations

DOI: 10.18421/tem81-16

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of accurately predicting used car prices in Bosnia and Herzegovina, a task complicated by the numerous distinct attributes influencing vehicle value. Motivated by the growing number of registered vehicles and the need for reliable pricing tools, the authors aim to develop a robust prediction model that outperforms traditional single-algorithm approaches. The research focuses on determining whether an ensemble of machine learning techniques can provide higher precision than individual classifiers, thereby assisting potential buyers and sellers in estimating fair market values. The methodology involved collecting data from the local web portal *autopijaca.ba* using a custom PHP web scraper. The initial dataset comprised 1,105 samples with attributes including brand, model, fuel type, mileage, power, and various feature flags (e.g., navigation, leather seats). After preprocessing to remove sparse attributes and outliers (such as brands with fewer than 10 samples or prices exceeding 60,000 BAM), the dataset was reduced to 797 samples. Continuous variables like mileage and price were converted into categorical intervals, transforming the problem from regression to classification. The authors evaluated three algorithms—Artificial Neural Network (ANN), Support Vector Machine (SVM), and Random Forest (RF)—both individually and as an ensemble. The data was split into 90% training and 10% testing subsets, with models built using the WEKA software. Initial tests using single classifiers yielded poor performance, with accuracies ranging from 41.18% for RF to 48.23% for SVM. Consequently, the authors proposed an ensemble approach where the dataset was divided into three price categories: cheap, moderate, and expensive. RF was used to classify cars into these broad categories. Subsequently, SVM and ANN were applied to each specific price subset to refine predictions. For the 90% split evaluation, SVM achieved the highest accuracy for cheap (86.96%) and expensive (90.48%) cars, while ANN performed best for moderate cars (86.11%). The final integrated model, which combines these specialized classifiers, achieved an overall accuracy of 87.38% on the test data. The system was implemented as a Java Swing GUI application to allow users to estimate car prices based on input features. The study concludes that combining multiple machine learning classifiers significantly enhances prediction accuracy compared to single-algorithm methods, which struggled to handle the complexity of the dataset. While the ensemble model requires more computational resources, it provides a reliable tool for price estimation. The authors suggest future work should validate the model against larger, international datasets from platforms like eBay and OLX to assess its generalizability beyond the local Bosnian market.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction