Detection of Shilling Attack with Support Vector Machines Using Oversampling

Recommender systems (RS) enable predicting the future behaviour of similar users by evaluating their past behaviour. Recommender systems overcome information overload problems. Collaborative Filtering (CF) provides quick recommendations for big databases. However, these models are vulnerable to shilling or profile injection attacks. Shilling attacks damage recommender system mechanism with synthetically generated fake profiles. These attacks aim to increase or decrease the rate of target item considering their attack types. In the current literature, enormous shilling attack detection methods are proposed. This study aims to detect future shilling attack detection with a classification approach. Thus, the robustness of RS is provided to future similar shilling attacks. This study uses Support Vector Machines (SVM) to robust RS models. However, attack sizes are very small compared to the dataset. A recently proposed Outlier Detection-Based Oversampling Technique (ODBOT) is used to solve this class imbalance problem. Comparative results are given for with ODBOT and without ODBOT models. Precision, recall, F1 measure and accuracy rates are presented for both models. The results show that the SVM model with ODBOT outperforms most of the performance metrics.