An Advanced Fusion Neural Network Paradigm for Intelligent Cyber Security Anomaly Detection

ABSTRACT Network security anomaly detection constitutes a critical defense mechanism in contemporary cybersecurity frameworks. We present a fusion CNN‐RNN (convolutional neural network‐recurrent neural network) model integrating spatial pattern recognition with temporal modelling for network anomaly detection, employing three‐branch architecture processing 41 attributes with regularisation for severe imbalance (U2R 0.1%). Comprehensive validation on the NSL‐KDD benchmark dataset establishes that our fusion paradigm outperforms existing machine learning approaches in both dual‐class and quintuple‐class classification challenges, achieving performance improvements of 11.5% and 29.8%, respectively.