Filter-based Attribute Selection Approach for Intrusion Detection using k-Means Clustering and Sequential Minimal Optimization Techniq

Intrusion Detection Systems (IDS) are administered by analysts for analysing system logs or data packets to predict malware in the network traffic. IDS automate this process for continuously increasing data in the network by using techniques based on machine learning and artificial intelligence, enabling packet detection without much human effort or intervention. Using machine learning to develop such systems can greatly influence delivering accurate and faster results with complex and intricate system logs and providing human-like responses, which have become are crucial in the real- world scenario today. In this paper, we propose a hybrid model using Filter-based Attribute Selection for reduction of feature dimension of the dataset. K-Means Clustering and Sequential Minimal Optimization (SMO) Technique of classification and machine learning are used for detecting various categories of attacks using KDD99 dataset for training and testing of the model. Our model is compared with other models for Intrusion Detection using various Performance metrics. Our proposed model provides prominent improvement in accuracy in detection compared to K-Means SVM model. The model also provides significantly higher correctly classified instances, lower incorrectly classified instances with consequential mean absolute error when compared to DBSCAN, K-Means++ & SMO model with regard to lower number of total instances.