An Ensemble-based Shill Billing Prediction Model in Car Auction System

The electronic auction system has emerged as one of the leading electronic commerce platforms where auctioneers and bidders converge for transactions. With the Internet's proliferation, e-commerce systems' functionalities have greatly been enhanced. Unfortunately, fraudulent activities increasingly hamper the credibility of online auction systems. Shill Bidding is one of the prominent frauds in the e-auction. Due to its similarity with normal bidding behaviors, it is challenging to detect as legitimate bidders could be categorized as fraudulent and vice versa. Several authentic auctioneers have been cheated during online bidding systems because of the diverse ways shill bidding is being perpetrated. It is, therefore, essential to improve the credibility of online bidding systems. In this study, we proposed a machine learning-based prediction system that determines the likelihood of a customer/seller perpetrating shill bidding. Upon deployment, the proposed system would prevent shill bidders from participating in a car action system. A vote ensemble model is trained with public data of 12 attributes comprising Random Forest, Decision Tree, Multi-layer Perceptron (MLP), and Sequential Maximal Optimization (SMO) base learners. An object-oriented Python programming language is used to implement the shill bidding predictive system. Experimental results show the excellence of the proposed system using metrics such as Precision, Accuracy, Recall, F1-score, and Misclassification error.