Algorithmic Approach for Fraudulent Transaction Detection using Market Basket Analysis with Big Data

Financial institutions that want to improve client experience alongside reduced financial losses need to identify transaction risks intensely. This research develops an integrated scheme for detecting retail fraud through joint utilization of FP-Growth clustering algorithms and ML models. Analysis requires preprocessing of Online Retail II UCI dataset for relevant feature extraction and Customer Engagement Index (CEI) assessment of customer behavior. The algorithms GMM, OPTICS, and DBSCAN detect normal and abnormal customer behaviors, while PCA enables easier visualization through dimensionality reduction. The analysis included Linear Discriminant Analysis (LDA) and Gaussian Naive Bayes (GNB) models, which demonstrated strong performance through GNB with 96.27% accuracy and 96.49% precision as well as 96.27% recall and 96.26% F1-score. The comparative analysis shows SVM, CNN and DT model shows lower accuracy in comparison to proposed models. Real-time retail fraud detection through recommended method delivered superior results than baseline models which proved its effectiveness and stability.