Hybrid Feature Selection for IoMT Based Intrusion Detection System For Integrating Mutual Information Filtering with Deep Learning Based Accelerated Metaheuristic Optimization

Hybrid Feature Selection Framework for IoMT Intrusion Detection Using Mutual Information Filtering and Deep Learning–Enhanced Dragonfly Optimization This repository contains the complete implementation of the research work titled: “Hybrid Feature Selection for IoMT-Based Intrusion Detection System Integrating Mutual Information Filtering with Deep Learning-Based Accelerated Metaheuristic Optimization.” The project proposes a two-stage hybrid feature selection framework designed to improve detection performance and computational efficiency in Internet of Medical Things (IoMT) intrusion detection systems. The framework integrates: Mutual Information (MI) FilteringA statistical filtering stage that measures the dependency between network traffic features and attack labels. Features are ranked according to their MI scores and redundant features are removed using correlation-based pruning. Deep Learning–Enhanced Binary Dragonfly Algorithm (DL-BDA)A wrapper optimization stage that refines the filtered feature space using a population-based metaheuristic inspired by dragonfly swarm behavior. The algorithm incorporates: Separation Alignment Cohesion Attraction to food source Enemy avoidance To reduce computational cost during wrapper optimization, a lightweight deep neural surrogate model (ProxyNet) is used to estimate the classification performance of candidate feature subsets. The proposed approach aims to simultaneously: maximize intrusion detection performance minimize feature redundancy reduce inference latency maintain compatibility with resource-constrained IoMT environments.