Year: 2025
Type: article
Abstract: Fraud detection in e-commerce has grown in importance as the volume of online transactions continues to rise. The rise in fraudulent behavior has led to significant financial losses and a decline in customer trust. This study explores the application of machine learning algorithms to identify fraudulent transactions, with a focus on anomaly detection methods. We examine many classification models, including XGBoost, Decision Tree, Random Forest, Bernoulli Naïve Bayes, and Logistic Regression, using a publicly available e-commerce fraud dataset. A range of performance criteria, such as the confusion matrix, F1-score, recall, accuracy, and precision, are used to assess the models. Random Forest achieved the highest accuracy (96.51%) of all the models tested, followed by XGBoost (95.22%) and Decision Tree (94.38%). With Optuna, Random Forest's accuracy was hyperparameter tuned to 97.08%. The results demonstrate the effectiveness of machine learning in detecting fraudulent transactions, with Random Forest emerging as the most dependable model. In addition to providing insights into improving fraud detection systems for e-commerce platforms, this research has the potential to inform future efforts aimed at improving model performance and real-time detection capabilities.
Authors: Ganesh SP, Aniruddha Nagesh Salvankar, Shawn Glanal Saldanha, Varun MC, Maryjo M George
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4410616819