Efficient Anomaly Detection in IoT Networks Using Logistic Regression and SMOTE

This research explores the application of the Logistic Regression Classifier for anomaly detection in IoT networks, utilizing the RT-IoT2022 dataset. The study focuses on identifying normal and anomalous network traffic across 12 classes, including diverse attack types such as DOS_SYN_Hping, NMAP_TCP_scan, and Metasploit Brute Force SSH. Comprehensive data preprocessing is performed to clean and normalize the dataset, and the Synthetic Minority Oversampling Technique (SMOTE) is applied to address class imbalance, improving the model’s ability to detect rare attack scenarios. The evaluation uses metrics such as precision, recall, F1-score, and area under the ROC curve (AUC). High performance is observed for most classes, with several achieving precision, recall, and F1-scores near 1.00, and AUC values exceeding 0.98, indicating strong classification accuracy. However, challenges are noted for certain classes, such as Metasploit Brute Force SSH and Wipro bulb, which exhibit higher rates of misclassification, as highlighted by the confusion matrix and error analysis. The model demonstrates consistent accuracy across cross-validation and test datasets, with a test accuracy of 0.98 validating its generalization capability. The results emphasize the effectiveness of Logistic Regression for anomaly detection while identifying areas for improvement in handling complex or underrepresented traffic patterns. Future work could focus on integrating advanced machine learning models, feature engineering, and real-time deployment to enhance detection accuracy and scalability in IoT networks. This study provides a robust foundation for developing efficient anomaly detection systems, contributing to improved security in IoT environments.