A metaheuristic-based ensemble feature selection framework for cyber threat detection in IoT-enabled networks

Abstract

Internet of Things (IoT) enabled networks are highly vulnerable to cyber threats due to insecure wireless communication, resource constraint architecture, different types of IoT devices, and a high volume of sensor data being transported across the network. Therefore, IoT-compatible cybersecurity solutions are required. An intrusion detection system is one of the most common solutions for detecting cyber threats in IoT-enabled networks. However, most of the existing solutions for cyber threat detection suffer from many issues like poor accuracy, high learning complexity, low scalability, and high false positive rate (FPR). We propose a metaheuristic-based intelligent and novel framework for cyber threat detection using ensemble feature selection and classification approaches to overcome these issues. First, a metaheuristic-based ensemble feature selection framework is designed using Binary Gravitational Search Algorithm (BGSA) and Binary Grey Wolf Optimization (BGWO) to get an optimized set of features to avoid the curse of dimensionality for efficient learning. Next, Decision Tree and ensemble learning-based classification techniques such as AdaBoost and Random Forest (RF) are employed separately to detect and classify cyber threats. The UNSW-NB15 dataset assesses the effectiveness of the proposed framework, and its performance is evaluated against recent state-of-the-art frameworks. Based on the result analysis, it is found that the RF outperforms existing modern cyber threats detection methods due to the optimized feature subset (4 features out of 42), maximum accuracy (99.41%), maximum detection rate (99.09%), and maximum F1-score (99.33%) with the lowest FPR (0.03%).