Comparative analysis of SVM Kernels and Parameters for Efficient Anomaly Detection in IoT

With increasing use of IOT in today's scenario, securing the network communication through anomaly detection becomes utmost important. ANN and Machine Learning technique like SVM have been used in the past also. However, existing solutions through SVM rarely compared the various variants and parameters for optimal performance of anomaly detection system. In this paper, we propose to find optimal parameters by using and comparing different kernel functions to implement SVM classifier effectively. Specifically, we used the NSL-KDD dataset in highly imbalanced form to test the model for robustness. Experiments validate that the model of SVM with optimum parameters can attain 99.9% on NSL-KDD dataset. Furthermore, it compares better to other methods in terms of accuracy and other metrics.