RADAR: Robust Anomaly Detection and Recognition for IoT Botnet Attacks Using Hyper-Dimensional Models

As our world continues to embrace digital technology, cyber threats have become increasingly prevalent, especially botnet attacks. Ensuring cybersecurity is a significant concern, and to address this, we need Intrusion Detection Systems (IDS) that can automatically adapt to new threats using advanced machine learning (ML) and deep learning (DL) techniques. Our project aims to develop an advanced anomaly detection and recognition system for IoT botnet attacks using hyper-dimensional models namely Random Forest, Decision Tree, Gradient Boosting, Multi-layer Perceptron (MLP), and K-Nearest Neighbor (KNN) and makes uses of hyperparameter tuning for performance optimization. By integrating an ensemble voting classifier that combines these five algorithms using a soft-voting mechanism, our results indicate an improved detection accuracy and adaptability. The proposed ensemble-based model is tested on a publicly available dataset which offers comprehensive coverage of botnet attack patterns and outperforms all individual classifiers, achieving an accuracy of 99.88% and an ROC AUC score of 0.999. The proposed approach highlights the advantages of combining hyper-dimensional classification models in effectively detecting IoT botnet attacks.