International Journal of Advanced Research in Computer and Communication Engineering

Abstract: The increasing deployment of IoT devices has heightened the need for effective security mechanisms to identify malicious activities within network traffic. With the rapid growth of IoT devices, safeguarding networks against malicious traffic has become increasingly critical. This study develops an intrusion detection system (IDS) using the UNSW-NB15 dataset, applying both supervised and unsupervised learning techniques. The dataset was preprocessed through feature selection, encoding, and cleaning, followed by exploratory analysis to reveal class imbalance and key traffic characteristics. A Long Short-Term Memory (LSTM) model was trained for binary classification of normal versus attack traffic, while a Bayesian Gaussian Mixture Model (BMM) was applied for anomaly detection using normal data. Evaluation employed accuracy, precision, recall, F1-score, ROC curves, and Youden’s Index for optimal threshold selection. Results show the LSTM delivered strong classification performance, while the BMM provided effective anomaly detection when thresholds were optimized. These findings highlight the potential of combining deep learning and probabilistic models to enhance IDS performance and strengthen network security.

Keywords: Intrusion Detection System (IDS), IoT Security, Network Traffic Analysis, UNSW-NB15 Dataset, Long Short-Term Memory (LSTM), Bayesian Gaussian Mixture Model (BMM),Anomaly Detection

Downloads: | DOI: 10.17148/IJARCCE.2025.141121