A Hybrid System of Deep Learning and Learning Classifier System for Database Intrusion Detection

Abstract

Nowadays, as most of the companies and organizations rely on the database to safeguard sensitive data, it is required to guarantee the strong protection of the data. Intrusion detection system (IDS) can be an important component of the strong security framework, and the machine learning approach with adaptation capability has a great advantage for this system. In this paper, we propose a hybrid system of convolutional neural network (CNN) and learning classifier system (LCS) for IDS, called Convolutional Neural-Learning Classifier System (CN-LCS). CNN, one of the deep learning methods for image and pattern classification, classifies the queries by modeling normal behaviors of database. LCS, one of the adapted heuristic search algorithms based on genetic algorithm, discovers new rules to detect abnormal behaviors to supplement the CNN. Experiments with TPC-E benchmark database show that CN-LCS yields the best classification accuracy compared to other state-of-the-art machine learning algorithms. Additional analysis by t-SNE algorithm reveals the common patterns among highly misclassified queries.