Use of OPTICS and Supervised Learning Methods for Database Intrusion Detection

Database security has become a prime concern in today's internet world due to the escalation of various web applications and information systems. Ensuring the security of the back-end databases is highly essential for maintaining the confidentiality and integrity of the stored sensitive information. In this paper, a Density-based clustering technique, namely, OPTICS, has been applied for constructing the normal profile of users. Each incoming transaction either lies within a cluster or is found to deviate from the clusters based on its Local Outlier Factor value. The transactions observed as outliers are further verified by employing various supervised machine learning techniques individually – Naïve Bayes, Decision Tree, Rule Induction, k-Nearest Neighbor and Radial Basis Function Network. The effectiveness of our system is demonstrated by carrying out extensive experimentations and comparative analysis using stochastic models.