Sparse Autoencoder with SGReLU Activation Based Network Intrusion Detection System

A Network Intrusion Detection System (NIDS) is the most important function, which is utilized to monitor malevolent activities frequently and detect these security attacks in Internet of Things (IoT) network. An improvement of effective performance and security is the most needed in the IDS system. This Sparse Autoencoder with Self-Gated Rectified Linear Unit (SGReLU) activation model for intrusion detection systems. The proposed Sparse Autoencoder with SGReLU activation function is evaluated using CIC-IDS-2017 and NSL-KDD dataset. The CIC-IDS-2017 and NSL-KDD dataset is standardized using Label encoder and one hot encoder method which improves the model performance. The pre-processed features are selected by using second percentile method and recursive feature elimination algorithms and the selected features are classified by using Sparse Autoencoder with SGReLU activation function which provides significant results. The proposed Sparse Autoencoder with SGReLU activation function model provides better results by utilizing evaluation metrics like accuracy, precision, f1-score values and PAR about 99.99%, 99.99%, 99.99% and 0.06 by using CIC-IDS-2017 dataset respectively when compared to the LSTM-CNN, NSOM, statistical method and LSTM-AE.