<![CDATA[The increasingly complex development of cyber threats has driven the need for network protection systems capable of detecting intrusions intelligently and automatically. Machine learning-based approaches have been widely adopted as intrusion detection solutions due to their ability to recognize attack patterns without relying on manually defined rules. This study examines the effectiveness of four machine learning algorithms, namely Random Forest, Decision Tree, Stacked LSTM, and AdaBoost, in network traffic classification tasks using the CICIDS2017 dataset as the evaluation benchmark. Before model training, the data was processed through normalization and relevant feature selection stages to ensure optimal data representation. Model performance was measured using four main evaluation metrics, namely accuracy, precision, recall, and F1-score, to obtain a comprehensive assessment of the classification capability of each algorithm. From the test results, Decision Tree stood out with a precision value reaching 99.87%, recall of 99.59%, and F1-score of 99.32%, indicating excellent per-class classification capability with a very low error rate. On the other hand, Random Forest recorded the highest overall accuracy value of 99.90%, accompanied by a precision of 97.78%, recall of 97.08%, and F1-score of 97.41%, reflecting the model's reliability in consistently maintaining performance across all classes. Both the Stacked LSTM and AdaBoost-based models showed lower performance compared to tree-based approaches in the same testing scenario. Overall, the results of this study confirm that decision tree-based algorithms offer a combination of high accuracy and superior classification stability, making them the right choice for implementing machine learning-based network intrusion detection systems.]]>
Copyrights © 2026
