Shilling attack detection in recommender system using PCA and SVM

Abstract

Shilling Attack is where, deceptive users insert fake profiles in recommender system to bias the rating, which is termed as shilling attack. Several studies were conducted in past decade to scrutinize different shilling attacks strategies and their countermeasures mainly categorized in linear algebra, statistical, and classification approach. This paper explores two different methods for shilling attack detection namely, Principal Component Analysis (PCA) and Support Vector Machine (SVM) and compares their performance on attack detection. We had experimented with simulating various attack models like average, random, and bandwagon models. This paper further discusses the importance of detecting malicious profile from the genuine one and suggests deep insights of developing new and more efficient shilling attack detection techniques. The experiments were conducted on the Movie Lens 100K Dataset and compared the performance of PCA technique with supervised SVM-classification method.