Real-time cybersecurity threat detection using AI-driven data mining and hybrid deep learning models
Abstract
The growing volume and complexity of cybersecurity data generated by Internet of Things (IoT) ecosystems require intelligent data mining techniques capable of extracting meaningful patterns and identifying malicious activities in real time. This paper proposes a data mining–driven cybersecurity framework that integrates Particle Swarm Optimization (PSO)-based feature mining with an attention-enhanced CNN-LSTM architecture for accurate threat detection in IoT environments. The proposed model considers the multifaceted spatial-temporal relationships of heterogeneous IoT network flows, which allows one to correctly detect anomalous and malicious processes. Particle Swarm Optimization (PSO) is used to optimize features and parameters in order to enhance the computational efficiency and scalability of the feature selection and parameter optimization process, dimensionality reduction of data, as well as retention of discriminative information. Moreover, model pruning and post-training quantization are added to reduce the computational cost and enable it to be deployed in resource-constrained IoT and cyber-physical settings. The framework is assessed on popular benchmark datasets, such as CICIDS2017, N-BaIoT, and IoT-23, that involve a diverse selection of cyberattack scenarios. The results of experiments indicate high detection, low false alarms, and low inference latency, which proves the effectiveness and real-time property of the proposed approach. The results indicate that combining deep learning and computational intelligence methods is effective in creating scalable and efficient security solutions for current distributed systems.