Velocity Paused Particle Swarm Optimization-based Intelligent Long Short-Term Memory Framework for Intrusion Detection System in Internet of Medical Things

Abstract

The rapid advancement of technology in the domain of Internet of Medical Things (IoMT) necessitates the implementation of stronger safety measures. Intrusion detection system (IDS) is an essential framework on IoMT networks proposed for identifying suspicious activities and detecting abnormal traffics. The implementation of the IDS framework might use either hardware or software solutions. However, traditional IDS frameworks are often inadequate in protecting data privacy and identifying complex and irregular intrusions specifically for IoMT environments. An improved velocity paused particle swarm optimization (VPPSO) methodology combined with long short-term memory (LSTM) has been proposed in this research to improve security in IoMT environment. The improved attack detection performance of proposed VPPSO_LSTM is achieved by the velocity pausing concept’s ability of VPPSO algorithm, which efficiently balance both exploration and exploitation. This enables an intelligent choice of hyper-parameter configurations for the LSTM architecture. This research has performed a comparative analysis to highlight the effectiveness of the proposed model with standard machine learning (ML) models including decision trees (DT), random forest (RF), adaptive boosting (AdaBoost), extreme gradient boosting (XGBoost), gradient boosting (GBoost), and CatBoost, as well as other deep learning (DL) methodologies such as LSTM, PSO_LSTM and Firefly Algorithm (FA)_LSTM. Implementing the ECU-IoHT dataset, the proposed VPPSO_LSTM model has been trained and validated for detecting and categorizing several attack types such as Smurf attack, ARP spoofing attack, Nmap port scans and denial-of-service (DoS) attack. For achieving better attack detection performance, the proposed model has been implemented with an optimal set of hyper-parameters including 64 LSTM cells in each LSTM layer, 32 neurons in each hidden layer, “ReLU” activation function for internal layers, dropout rate of 0.1, optimizer as “Adam” and a learning rate of 0. 001. The proposed VPPSO-LSTM approach shows improved performance by highlighting an accuracy score of 99.98%, ROC-AUC score of 0.9999, recall score of 0.9994, precision score of 0.9996 and an F1 score of 0.9995. The proposed VPPSO-LSTM approach makes a significant contribution to improving IoMT security with its optimal hyper-parameters setup and outperforming conventional ML and DL models. This research highlights opportunities for further investigation into efficient, scalable and real-time IDS specifically designed for IoMT environments.