EWCMD: real-time attack detection using ensemble weighted combination machine learning and deep learning methods

Abstract The evolving IoT threat landscape makes Distributed Denial-of-Service (DDoS) attacks a persistent security concern. This paper proposes a five-stage ensemble intrusion detection framework combining machine learning and deep learning. The first four stages train and evaluate seven machine learning classifiers and a convolutional neural network (CNN) on the NSL-KDD and UNSW-NB15 datasets, incorporating feature selection and hyperparameter optimization to improve accuracy and reduce computational cost. The fifth stage introduces two final prediction strategies: a weighted roulette-wheel mechanism (ERCMD) for low-latency decisions and a CNN-based meta-classifier (EDCMD) for maximum accuracy. ERCMD achieves over 96% accuracy on both datasets, while EDCMD reaches up to 100% accuracy on NSL-KDD and 99.95% on UNSW-NB15.