Abstract: The rapid growth of e-commerce has led to a rise in fraudulent interactions, causing disruptions to digital platforms and the financial ecosystems that support them. The study applied various classification models, including Naïve Bayes, Decision Tree, Random Forest, and K-Nearest Neighbors, and evaluated their performance using key metrics such as accuracy, precision, recall, and F1 score. Model selection was aided by information gain measures for attribute selection to enhance performance. Additionally, model accuracy was improved through hyperparameter tuning and ensemble methods, such as voting across models at the same risk level. Overall, the evaluation showed that classification models, especially well-tuned ones, are effective solutions for e-commerce fraud detection. This research also provides a foundation for developing automated fraud detection systems and supports the ongoing efforts to improve cybersecurity and trust in online commerce.

Keywords: E-Commerce Fraud, Fraud Detection, Imbalanced Classification, Data Mining, Machine Learning, Transaction Risk Analytics

Pages: 214-229