Intelligent Cyber Threat Detection In Cloud Computing Using Hybrid Genetic Feature Selection and CONV-BILSTM
Abstract
The rapid growth of cyber threats in cloud computing requires automated detection and incident-response pipelines that can process high-dimensional network traffic with low false-alarm rates. This paper presents an AI-driven intrusion-detection framework that couples Hybrid Genetic Algorithm Feature Selection (HGAFS-IDS) with a convolutional bidirectional long short-term memory (Conv-BiLSTM) classifier. The key contribution is the integration of evolutionary dimensionality reduction with joint spatial-temporal learning in a cloud-deployable inference pipeline. Network-flow records are cleaned, encoded, normalized, and reduced by HGAFS-IDS before classification by the Conv-BiLSTM model. The framework is evaluated on NSL-KDD, UNSW-NB15, and CIC-IDS-2017. In the reported model comparison, the proposed approach achieves 95.08% accuracy, 95.36% precision, 95.08% recall, and 95.22% F1H. Dataset-wise accuracy ranges from 89.75% to 95.20%. A Flask-based API demonstrates cloud integration for online inference and alert generation. The analysis is behavior-based and operates on dynamic network/flow telemetry rather than static executable-binary inspection. The results show improved point-estimate performance over the reported Random Forest and SVM baselines, while the study remains limited by benchmark age, the absence of archived repeated-run statistics, and the lack of measured high-volume multi-cloud latency and throughput.