An Efficient Probabilistic Supervised Learning Model For Realtime Iot Attack Detection in Software-Defined Iot (SD-Iot) Networks

Abstract

As the size of an imbalanced database increases, the true positive rate of traditional prediction algorithms faces difficulties in improvement due to the high ratio of majority to minority classes and the substantial noise present in the data. Missing values, attribute noise, and imbalanced classes are some of the significant factors that can impact the quality of input data. The quality of imbalanced data significantly impacts the efficiency of classification approaches, making it necessary to ensure high-quality input data to achieve optimal results. Therefore, to ensure high-quality predictions on imbalanced datasets, traditional machine learning models need to be optimized. In this paper, a filtering approach, feature ranking, clustering, and ensemble classification model are proposed to address these issues in imbalanced SD-IoT datasets. A novel strategy is proposed to handle missing data, imbalanced classes, feature selection, probabilistic clustering, and ensemble classification approaches are optimized to improve the true positive rate and error rate on imbalanced SD-IoT databases. This work implements a hybrid probabilistic class membership and optimized Max-Hellinger based classification framework on imbalanced SD-IoT datasets to improve overall statistical metrics. Experimental results have demonstrated that the proposed approach outperforms conventional privacy-preserving techniques in terms of statistical metrics.