An Efficient Intrusion Detection System for IoT Using Feature Engineering and Machine Learning

Abstract

IoT devices have limited computing and security capabilities, making them vulnerable to cyberattacks. This rapid expansion of IoT has introduced unprecedented connectivity but also heightened security vulnerabilities. The security of IoT environments, therefore, depends on efficient and lightweight intrusion detection systems (IDS). Using advanced feature engineering and machine learning algorithms, this study develops high-performance IDS designed for IoT networks. Data imbalance is addressed with preprocessing techniques, feature extraction, and synthetic minority oversampling techniques (SMOTE). Multi-dataset training and testing included K-nearest neighbour models, sequential minimalism optimization models, random forest models, and stacking ensembles. A transfer learning model such as VGG-16 and DenseNet was also incorporated to improve classification accuracy. It has been demonstrated that the proposed models, particularly the ensemble and RF-based approaches, are highly accurate, precise, and recallable. IoT environments with limited resources can benefit from the proposed IDS framework because it effectively identifies malicious traffic while maintaining computational efficiency.