Intrusion Detection and Prevention Using Machine Learning for IoT-Based
Rajesh, Mridul Chawla · International Journal of Computational and Experimental Science and Engineering · 2025
The intrusion detection system (IDS) is an essential component for enterprises as it safeguards network infrastructure, assets, and confidential data, effectively preventing cybercriminal actions. Various strategies have been devised and put into action in order to prevent malicious activity up to this point. Given the efficacy of machine learning (ML) techniques, the proposed strategy utilized multiple ML techniques for the IDS. The UNSW-NB15 dataset was utilized to conduct an offline analysis of models’ performance and to create a tailored integrated classification system for detecting malicious activities in a network. The performance analysis involved training and evaluating the “Decision Tree (DT)”, RF, CatBoost, and Hybrid models for a binary classification task. To address the decline in the performance of IDS while using a feature vector with a large number of dimensions, a Gini Impurity-based Weighted Random Forest (GIWRF) model was utilized to choose the most suitable set of features. This approach served as the incorporated choosing features technique. Additionally, feature extraction was performed using the Genetic algorithm (GA). This method utilized Gini impurity in order to enhance the learning algorithm’s comprehension of the class distribution. 27 features were chosen from UNSW-NB 15 based on their relevance value. The results of the study showed that the Hybrid model scored better than the other trained models used in the present study. This study offers useful insights on enhancing the security of IoT networks through the utilization of ML. The research also quantified various attacks (DOS, Probe etc.), assessing their detection efficiency using the hybrid model. The findings proved high accuracy in detecting various threats, further confirming the strength of the proposed method. Study highlights the significance of customized strategies and continuous improvements in increasing the resilience of systems to constantly changing cyber-attacks. In addition, the GIWRF-Hybrid method proposed in the paper showed better performance than other methods considered in the paper, that is, accuracy and loss.