A Hybrid Multilayer Perceptron and Firefly Optimization for Network Intrusion Detection System

The increasing sophistication and frequency of cyber-attacks have made the development of the effective [IDS] more critical than ever. This challenge is further amplified by the growing deployment of Unmanned Aerial Vehicles (UAVs). In response to these challenges, this paper proposes an advanced IDS model that is designed for UAV detection which leverages a Multi-Layer Perceptron (MLP) architecture, specifically the model is further enhanced by a hybrid optimization strategy that integrate the Firefly Algorithm with MLP to improve classification accuracy and detection performance. The approach proposed begins with comprehensive data preprocessing, where raw binary network traffic data is converted into a CSV format to streamline model input. To emphasize the issue of class inequality commonly found in UAV Cyber Security datasets, approach strategy an oversampling technique is applied ensuring the model can effectively learn from both attack and normal UAV traffic patterns. Additionally, data normalization techniques are employed to standardize inputs, optimizing the performance of the model during development. This dataset array is partitioned into 80% for training and 20% for testing, enabling a rigorous evaluation process. The model utilizes the ReLU response function for the hidden layers and soft-max for the output layer, focusing on binary classification to distinguish between attack and normal UAV traffic. The functioning of the model is analyzed accessed using key quantifiers, including exactness, precision, recognition, and F1-score, along the results indicating that the proposed IDS achieves an accuracy of 89%, overtaking the execution of prevalent Machine Learning (ML) and Deep Learning (DL) - based IDS techniques in UAV threat detection.