Mitigating malicious denial of wallet attack using attribute reduction with deep learning approach for serverless computing on next generation applications.

Denial of Wallet (DoW) attacks are one kind of cyberattack whose goal is to develop and expand the financial sources of a group by causing extreme costs in their serverless computing or cloud environments. These threats are chiefly related to serverless structures owing to their features, such as auto-scaling, pay-as-you-go method, cost amplification, and limited control. Serverless computing, Function-as-a-Service (FaaS), is a cloud computing (CC) system that permits developers to construct and run applications without a conventional server substructure. The deep learning (DL) model, a part of the machine learning (ML) technique, has developed as an effectual device in cybersecurity, permitting more effectual recognition of anomalous behaviour and classifying patterns indicative of threats. This study proposes a Mitigating Malicious Denial of Wallet Attack using Attribute Reduction with Deep Learning (MMDoWA-ARDL) approach for serverless computing on next-generation applications. The primary purpose of the MMDoWA-ARDL approach is to propose a novel framework that effectively detects and mitigates malicious attacks in serverless environments using an advanced deep-learning model. Initially, the presented MMDoWA-ARDL model applies data pre-processing using Z-score normalization to transform input data into a valid format. Furthermore, the feature selection process-based cuckoo search optimization (CSO) model efficiently identifies the most impactful attributes related to potential malicious activity. For the DoW attack mitigation process, the bi-directional long short-term memory multi-head self-attention network (BMNet) method is employed. Finally, the hyperparameter tuning is accomplished by implementing the secretary bird optimizer algorithm (SBOA) method to enhance the classification outcomes of the BMNet model. A wide-ranging experimental investigation uses a benchmark dataset to exhibit the superior performance of the proposed MMDoWA-ARDL technique. The comparison study of the MMDoWA-ARDL model portrayed a superior accuracy value of 99.39% over existing techniques.