Predictive Analytics for E-Commerce Fraud Detection Using Machine Learning Algorithms

Abstract

The rapid development of e-commerce provides new challenges faced by society, one of them is fraudulent e-commerce transactions. Losses caused by e-commerce fraud globally are expected to exceed 48 billion USD by 2023. The use of advanced technology such as machine learning can be a solution in an effort to detect and prevent e-commerce fraud. This research aims to evaluate several machine learning algorithms, such as deep learning, naive bayes, logistic regression, decision tree, and neural network, to detect e-commerce fraud. The dataset used consists of 1,472,952 transactions. This research consists of several stages, namely: data retrieval, weighting, feature selection, normalization, data sharing and data analysis. At the analysis stage, the algorithms were compared using a confusion matrix consisting of sensitivity, precision, accuracy, and F1 Score. The results show that each algorithm used gets a very high test value with a percentage of more than 90%.