Optimization Of Intrusion Detection Using Likely Point PSO And Enhanced LSTM-RNN Hybrid Technique In Communication Networks

Abstract

“Intrusion detection system (IDS)” is a significant component of maintaining secure communication networks and all network managers have embraced it to their happiness.  Several methods have been proposed on the early intrusion detection systems.  Nevertheless, they possess issues that render them less efficient in dealing with new/different threats in the future.  To make IDS more secure we propose the application of the “enhanced long-short term memory (ELSTM) approach with a recurrent neural network (RNN) (ELSTM-RNN)”.  Intrusion detection systems have had several associated challenges which include gradient vanishing, generalization, and overfitting.  In the proposed method, the issue of gradient clipping is addressed “through probably point particle swarm optimization (LPPSO) and enhanced LSTM classification.  To test and validate the proposed method, we have used NSL-KDD dataset (KDD test PLUS and KDD TEST21)”.  particle swarm optimization is a superior technique we used to select numerous helpful features.  The selected attributes are utilized to classification with high performance using an enhanced LSTM model, which is adopted to rapidly classify attack data among the normal data.  To retest the proposed system, “we applied it to the u.s.-NB15, CICIDS2017, CSE-CIC-IDS2018, and BOT_DATASET datasets suggested”.  The findings indicate that the proposed system requires less time to train compared with current methods of various classes.  “Finally, the performance of the proposed ELSTM-RNN architecture is examined with the help of several metrics such as accuracy, precision, recall, and error rate”.  Our approach outperformed DNNs approaches.

 we try out an ensemble way to improve performance, using a voting Classifier using voting classifier and stacking

classifier algorithms. This method achieves an amazing accuracy of 100%.  This ensemble method combines several distinct models to make an intrusion detection system that is more stable and dependable.  The results show that the suggested ELSTM-RNN architecture works well and might be improved even more by using ensemble methods. this is a huge step forward for IDS security and performance.