Firefly Algorithm-Optimized Deep Learning Model for Cyber Intrusion Detection in Wireless Sensor Networks Using SMOTE-Tomek

Abstract

Wireless Sensor Networks are increasingly vulnerable to sophisticated cyber threats, necessitating effective and intelligent intrusion detection strategies. This paper presents a deep learning-based intrusion detection model that enhances cybersecurity performance through intelligent hyperparameter optimization and advanced data balancing. The main objective of the study is to improve classification accuracy and generalization in intrusion detection systems by employing a dynamic and adaptive optimization framework. A Firefly Algorithm that mimics nature is part of the suggested model. It changes the number of neurons, learning rate, and dropout rate while evaluating the performance of every arrangement with just a little training. It uses the strategy of swarms to find solutions effectively and in an adaptable way. A hybrid method called SMOTE-Tomek is employed to deal with the issues caused by an unequal number of classes in the dataset. The network is built with different dense layers that are enhanced with dropout and batch normalization, and adaptive learning rate adjustment. In preprocessing the data, we encoded categorical variables, made the values consistent with normalization, and balanced the classes by producing artificial data when needed. The model was trained using GPU software for ten epochs and checked for performance using accuracy measurements, confusion matrices, and classification reports. The optimized model obtained an accuracy of 97.82% in classifications, higher than what baseline models and previous machine learning methods could do. It is able to spot and classify various kinds of attacks, completely handles Flooding cases and greatly lowers the chances of mistakes when identifying Blackhole and Grayhole. The study underlines the fact that using swarm intelligence with hybrid resampling enhances the real-time protection of networks against cyber attacks. A deep learning framework is developed at the end of the study that can work well and effectively in cybersecurity tasks.