MODELLING A DEEP LEARNING BASED HYBRIDIZED ENSEMBLE LEARNING APPROACH FOR INTRUSION DETECTION IN INDUSTRIAL ENVIRONMENT

Abstract

A network intrusion detection system (NIDS) is one of the primary approaches for preventing cyber-attacks. Deep learning (DL) has shown promising results for intrusion detection, yet the effect of feature fusion on improving accuracy and generalization in NIDS remains underexplored. This study proposes novel deep learning frameworks employing feature fusion strategies to enhance the reliability of multi-class classification in NIDS. This work introduces three distinct approaches based on Ensemble Convolutional Long Short-Term Memory (Ensemble C LSTM) models that integrate multiple feature types to address class imbalance and capture inter-feature correlations effectively. Publicly available datasets, UNSW-NB15 and NSL-KDD, were used to evaluate the proposed models against existing algorithms. Experimental results demonstrate that the late-ensemble and late fusion approaches achieved superior generalization, with consistent performance on validation and test sets, while minimizing overfitting. Compared to conventional DL and modern algorithms, the proposed frameworks deliver improved robustness in handling class imbalance and complex classification tasks. These results indicate that feature-fusion-based deep learning models offer a promising pathway toward building reliable intrusion detection systems capable of addressing evolving cyber threats.