LDA-OOBO Based Dimensionality Reduction and Classification Using Hybrid BiGRU-MLP for Web Based Cyber-Attack Prediction in Industrial System

Cyberattacks have increased, posing significant threats to economies, enterprises, infrastructure and as well as to individual users. More than 430 million new, distinct malware components were found by Symantec. Since cyber-security threats are becoming more diverse and unexpected, traditional cyber-attack detection systems are facing several serious challenges as they process massive amounts of data. These challenges include inappropriate feature extraction and selection, lengthy prediction computation times, and erroneous classification models. In this paper, study of web-based cyberattack detection in industrial systems has been done. The proposed architecture gathers and compiles the website log files into a dataset. These gathered datasets are pre-processed using Generative Adversarial Imputation Net (GAIN) to impute the missing values in the raw dataset. Next, they are examined to choose characteristics from the dataset using the LDA with one-to-one based optimization technique. Anticipated the cyberattack by applying the Bilateral-Gated Recurrent Unit Network (BIGRU-MLP) approach. Evaluated Performance indicators like Accuracy, Error, Precision and recall values are 98.91%, 1.09%, 99.01% and 98.80%. The results clearly shows that the proposed BiGRU-MLP approach is one of the optimal approaches for Web-based cyberattack detection in industrial systems.