Predictive Models for Cybersecurity in Smart Cities Network Using NSL-KDD Dataset

Abstract

Smart Cities increasingly rely on interconnected digital infrastructures and Internet of Things (IoT) systems, which expand the attack surface and create new cybersecurity challenges. Traditional intrusion detection systems (IDS) based on signatures and rules are limited in scalability and adaptability against zero-day and polymorphic threats. This study presents a comparative evaluation of three supervised machine learning models, Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), and Artificial Neural Network (ANN), for intrusion detection using the NSL-KDD dataset. A rigorous preprocessing pipeline, including feature encoding, normalization, and stratified sampling, was applied to ensure balanced and reliable training and testing. Models were evaluated on accuracy, precision, recall, and F1-score, metrics critical for real-world deployment. Experimental results show that ANN achieved the highest performance (accuracy: 98.19%, precision: 99.97%, recall: 99.19%, F1-score: 99.77%), followed by SVM with a strong balance between detection and efficiency, while LDA provided lightweight and interpretable outcomes suitable for constrained edge environments. The findings highlight the trade-offs between accuracy, computational cost, and interpretability, offering actionable insights for selecting intrusion detection models tailored to Smart City cybersecurity infrastructures. This work advances machine learning-based IDS strategies by addressing performance and practical deployment considerations in large-scale, heterogeneous, and latency-sensitive networks.