Deep Belief Network Based Intrusion Detection System with L 1 and L2 in IoT Environment

Intrusion detection plays a crucial role in identifying unknown attacks from network traffic in Internet of Things environment and is an essential component of modern cybersecurity. Traditional machine learning models such as XGBoost and Random Forest have been widely applied for network anomaly detection. However, these methods suffer from relatively low accuracy and rely heavily on handcrafted feature engineering, which is inadequate in the era of big data. To address these challenges, deep learning-based models, such as Deep Belief Networks (DBN) has been introduced to improve feature extraction. Despite their advancements, existing models still struggle with high-dimensional network traffic data and feature redundancy. In this study, we propose a novel DBN-L1 and DBN-L2-based Intrusion Detection System (IDS) that significantly enhances anomaly detection accuracy. DBN-L1 and L2 leverage layer-wise unsupervised pretraining followed by fine-tuning to learn hierarchical representations of network traffic data. The design of these models allows them to automatically discern critical traffic features, thereby eliminating laborious manual selection and simultaneously enhancing their classification accuracy. To assess our method, we have utilized the NSL-KDD dataset for Internet of Things networks, a key reference point for validating intrusion detection systems. Deep learning proves highly effective in enhancing network security and providing precise anomaly detection.