DDoSNet: Detection and prediction of DDoS attacks from realistic multidimensional dataset in IoT network environment

Abstract

The Internet of Things (IoT) network infrastructures are becoming more susceptible to distributed denial of service (DDoS) attacks because of the proliferation of IoT devices. Detecting and predicting such attacks in this complex and dynamic environment requires specialized techniques. This study presents an approach to detecting and predicting DDoS attacks from a realistic multidimensional dataset specifically tailored to IoT network environments, named DDoSNet. At the beginning of the data preprocessing phase, the dataset must be cleaned up, missing values must be handled, and the data needs to be transformed into an acceptable format for analysis. Several preprocessing approaches, including data-cleaning algorithms and imputation methods, are used to improve the accuracy and dependability of the data. Following this, feature selection uses the African Buffalo Optimization with Decision Tree (ABO-DT) method. This nature-inspired metaheuristic algorithm imitates the behaviour of African buffalos to determine which traits are the most important. By integrating ABO with the decision tree, a subset of features is selected that maximizes the discrimination between regular network traffic and DDoS attacks. After feature selection, an echo-state network (ESN) classifier is employed for detection and prediction. A recurrent neural network (RNN) that has shown potential for managing time-series data is known as an ESN. The ESN classifier utilizes the selected features to learn the underlying patterns and dynamics of network traffic, enabling accurate identification of DDoS attacks. Based on the simulations, the proposed DDOSNet had an accuracy of 98.98 %, a sensitivity of 98.62 %, a specificity of 98.85 %, an F-measure of 98.86 %, a precision of 98.27 %, an MCC of 98.95 %, a Dice coefficient of 98.04 %, and a Jaccard coefficient of 98.09 %, which are better than the current best methods.