Anomaly Detection in Real-Time IoT Systems Using Machine Learning

The increase of Internet of Things (IoT) systems has revolutionized industries such as industrial control, smart city, and healthcare monitoring systems, among others, by offering real-time data sharing and decision making. Nevertheless, IoT systems are subjected to severe operational challenges—such as faults, cyber-attacks, and bugs—that may impede their normal execution. Conventional anomaly detection techniques such as rule-based systems and statistical models fall short of handling the complex and dynamic nature of modern-day IoT systems. The study proposes an unsupervised method for real-time anomaly detection with an application to NSL-KDD dataset in the domain of intrusion detection using a stacked autoencoder (SA) and principal component analysis (PCA) for feature extraction and dimensionality reduction. The classification approach is based on Random Forest classifier with Random Under Sampler and class-weight balancing to deal with class imbalances. The combination approach is a scalable approach which is adaptive, efficient enough for computation and can be deployed on resource-constrained IoT environments. The experimental results demonstrate that the approach is superior to the existing methods with high detection accuracy and low false positives, which is suitable for protecting IoT environments to assist them with high-performance operations.