부산대학교 도서관

The Internet of Things (IoT) represents a transformative technology with profound implications for our daily lives. However, the rapid proliferation of IoT devices has brought about heightened concerns regarding security vulnerabilities. To address these challenges, this research focuses on the application of Machine Learning (ML) algorithms to detect anomalies in the CTU-IoT dataset, a prominent IoT-based home automation dataset. The methodology entails extracting comma-separated values (.csv) files from packet capture (pcap) data, followed by the removal of outliers using a simple imputer technique. The Isolation Forest algorithm is then utilized to label dataset entries as anomalies or not, with these labels incorporated into the dataset. Given the class imbalance issue that arises, oversampling techniques such as SMOTE (Synthetic Minority Oversampling Technique) and Borderline-SMOTE are employed to balance the dataset. The study encompasses three datasets: the original unbalanced dataset, the SMOTE balanced dataset, and the Borderline-SMOTE dataset. A selection of ML algorithms, including SVM, Naive Bayes, Decision Tree, and Random Forest is applied to these datasets, and their performance is assessed using key metrics like accuracy, precision, recall, F1-score and computation time. The findings reveal that the Decision Tree algorithm outperforms the other three algorithms based on the performance metrics, showcasing its suitability for anomaly detection in the Borderline- SMOTE balanced dataset. This research contributes to strengthening the security and safety of IoT applications.