ASBO-MAT-BiLSTM: Social Media Network Anomaly Detection Using Optimized Multihead Attention Transformer based Model

Currently, attacks in the networks are the most vital issue in modern society and the networks from minor to huge networks are susceptible to network threats. Various approaches exist with several advantages and limitations in detecting anomaly, such as computational complexity, overfitting, low level of accuracy in detection, and so on. To overcome these issues and detect network anomaly effectively Adaptive Secretary Bird Optimization algorithm (ASBO) is integrated with the RAnked Minority Oversampling in Boosting (RAMOB) model to resolve the issue of data imbalance through making the data in a balanced manner. The Multihead Attention Transformer (MAT) mechanism and ASBO algorithm enable the BiLSTM model to effectively detect anomaly by solving the issues of overfitting and complexities in terms of both time and computation. The experimental outcomes display that the ASBO-MAT-BiLSTM model reaches an accuracy of 97.116% using NSL-KDD Dataset and accuracy of 96.54%, using CIC-IDS2017 Dataset, which reveals that our model attains higher efficiency in detecting anomalies in the networks.