Abstract
The increasing deployment of Internet of Things (IoT) devices has made networks more vulnerable to malicious activities, necessitating effective traffic classification methods. This study investigates machine learning-based approaches to classify network traffic using the CTU-IoT-Malware-Capture-3-1conn.log.labeled dataset, which includes over 150,000 records labeled as benign or malicious. Key numeric features such as packet counts, data volumes (bytes), protocol types, and source IP frequencies were selected to enhance the models' predictive capabilities. Preprocessing involved normalization with StandardScaler to ensure equal contribution of features to model predictions. Four machine learning algorithms were evaluated: Logistic Regression, Random Forest,Support Vector Machines (SVM), and Gradient Boosting. The experiments utilized a 5-fold cross-validation framework to ensure robustness and reliability. Gradient Boosting outperformed other models with a mean accuracy of 99.13%, followed by Random Forest at 99.11%. To address class imbalance, SMOTE was applied, significantly improving the recall of minority classes. This study demonstrates the potential of machine learning for improving IoT network security by identifying anomalous behaviors in real-world attack scenarios. The results underline the importance of preprocessing, feature selection, and evaluation in achieving highdetection accuracy.