Feature selection and hybrid CNNF deep stacked autoencoder for botnet attack detection in IoT

Botnet attack is a severe cyber security issue, which occurs in the Internet of Things (IoT). These attacks are carried out by hackers to acquire control of various IoT devices and carry out illegal activities. Though several methods have been proposed to overcome these issues, the rapidly evolving nature of botnet makes attack detection complicated. Hence, in this paper a Deep Learning (DL) model is introduced for identifying botnets in IoT. Initially, the IoT network is simulated, and the detection of attack is established using the log data. Afterwards, the log data is fed into data pre-processing, in which the data is pre-processed by Quantile normalization. Then, feature selection is effectuated by employing Information Gain (IG), and City Block Distance. Once the feature selection is performed, data augmentation is done with the use of oversampling to increase the samples. Lastly, the Botnet attack detection is carried out by using the proposed Convolutional Neural Network Fused with Deep stacked Autoencoder (CNN-FDSA), which is formed by fusing Deep Stacked Autoencoder (DSA) and Convolutional Neural Network (CNN). Furthermore, the proposed CNN-FDSA attained the highest recall, precision, f-measure, and accuracy of 90.3 %, 91.6 %, 90.9 %, and 92.4 %, and then the lowest False Positive Rate (FPR) of 8.2 %.