Intrusion Detection System Using Computationally Efficient SVM based on K-Prototype Clustering
Bharath Kumar Reddy Athipalli, S. K. Gargee, Sai Ruthwik Kancharla, Krishna Prasad T R · 2022 Third International Conference on Intelligent Computing Instrumentation and Control Technologies (ICICICT) · 2022
Intrusion Detection Systems (IDS) are crucial to the safety of a network because it allows the IT personnel to be warned about suspicious activity and malicious attacks. Over the years, there has been a lot of improvement in IDS technology using Machine Learning Techniques. In this paper, we propose a new hybrid ML model for IDS using K-prototype clustering and Support Vector Machines (SVM). Support vector machines being a large margin classifier is very efficient in terms of accuracy for classification problems. However, when the training data is large, it can be computationally expensive and time-consuming. In this aspect, there is scope for improvement using the K-prototype model. The original KDD-1999 data set consists of 4.9 million rows and 43 columns. Training an unmodified SVM model on the entire data set can be very costly in terms of the number of computations and time taken to train the model. We designed a model that effectively chooses the most important samples of data for training SVM without compromising the model in terms of accuracy using a combination of K-Prototype and SVM. The proposed model reduces the number of training samples, which in turn reduces the number of support vectors thereby reducing the time taken to train SVM while still producing impressive results.