Efficient ML-Based Identification Models of Network Intrusion in Internet of Things (IoT) Systems

Abstract

The wide adoption of IoT devices has brought about increased connectedness and convenience, but it has also revealed new security concerns. Protecting IoT systems from malicious hackers is crucial as they become more important in various areas. Using the Random Forest (RF) method for machine learning, this study presents an efficient and effective Intrusion Detection System (IDS) tailored to Internet of Things (IoT) devices. The research employs a wide variety of data processing approaches, including data cleaning, normalization, and one-hot encoding, to get the NSL-KDD dataset ready for model training and testing. With a remarkable 99.9 % accuracy, precision, recall, and F1-score, the RF model surpassed other conventional modelling approaches, such as Decision Tree (DT), Support Vector Machine (SVM), and Conv1d networks. Using ensemble learning effectively is where this approach really shines. Interpretability was also facilitated by visualization (including correlation heatmaps and feature importance plots), which helped in selecting features. Effectiveness of the system under verification was confirmed with consistent evaluation metrics and confusion matrices analysis, and RF in the ability to reliably differentiate between malicious and normal traffic