Web Security Attack Detection Using Knowledge-aware Attentional Neural Network optimized with Dipper Throated Optimization Algorithms

Web applications are often susceptible to assaults since they store valuable assets and critical data. Numerous security problems concerning intrusion in both web applications and computer network frameworks have arisen due to the rapid expansion of web applications. Web clients and web servers can communicate. Suspicious changes to normal web requests are referred to as web assaults. The detection is performed rapidly and appropriately for effective operation of highly solicited web security. In this paper, a Web Security Attack Detection Using Knowledge-aware Attentional Neural Network optimized with Dipper Throated Optimization Algorithms (WSAD-KANN-DTOA) is proposed. Here, the input data is collected from CSE-CIC-IDS2018 dataset. After that, the input data are preprocessed with the help of Subaperture Keystone Transform Matched Filtering (SAKTMF) to normalize the data. Then, the preprocessed data undergo feature extraction by Directional Lifting Wavelet Transform (DLWT) to extract gray-scale statistical features like mean, skewness, standard deviation, homogeneity. Then the extracted data is fed to Knowledge-aware Attentional Neural Network (KANN) to categorize the web security attacks as Normal, DDOS, BOT, Inf, Brute-force, Dos and Web attack. Generally the KANN does not express some adaptation of optimisation methods to identify the optimum parameters to ensure proper categorization of online security attack detection as malicious or normal. Hence, Dipper Throated Optimization Algorithm (DTOA) is used to optimize KANN, which accurately classifies the web security attack detection. Then the proposed WSAD-KANN-DTOA approach is applied and effectiveness is examined utilizing performance metrics like accuracy, precision, sensitivity, F1-score, detection rate, ROC error rate. The performance of WSAD-KANN-DTOA approach attains 2.55%, 4.72% and 4.63% better accuracy; 8.66%, 8.97%, 7.57% better precision; 7.18%, 6.52%, 8.68% better sensitivity is compared with existing models: Novel Class Probability Features using Logistic Regression for Network Attack Detection (NCPF-LR-NAD), Cyber Threat Detection With BERT-Based Lightweight Model (CTD-BERT-BLM), Web Based Depend Anomaly Detection utilizing zero-shot learning with CNN(WBAD- CNN-WS) respectively.