Detecting Unbalanced Network Traffic Intrusions With Deep Learning

The growth of cyber threats demands a robust and adaptive intrusion detection system (IDS) capable of effectively recognizing malicious activities from network traffic. However, the existing imbalance of class in network data possesses a significant challenge to traditional IDS. To overcome these challenges, this project proposes a novel hybrid Intrusion Detection System using machine learning algorithms, which includes XGBoost, Long Short-Term Memory (LSTM), Mini-VGGNet, and AlexNet, which is used to handle the unbalanced network traffic data. Furthermore, the Random Forest Regressor is used to ascertain the importance of features for enhancing detection accuracy and interpretability. Addressing the inherent class imbalance in network data is crucial for ensuring the IDS's effectiveness. The proposed system employs a combination of oversampling techniques for minority classes and under sampling techniques for majority classes during data preprocessing. This balanced representation of network traffic data helps prevent the IDS from being biased towards the majority class and improves its ability to detect rare or novel intrusions. The utilization of Random Forest Regressor for feature extraction serves a dual purpose. It helps identify the most relevant features within the network traffic data that contribute significantly to detecting intrusions. It enables the system to prioritize and focus on these important features during model training, thereby enhancing detection accuracy while reducing computational complexity. This research contributes to the ongoing efforts to mitigate cyber threats and safeguard critical network infrastructures.