<![CDATA[Network intrusion has rapidly evolved, posing significant risks to IT infrastructure. To address this, ensemble learning, known for its robust classification capabilities, is applied to IoT network traffic using the public RT_IOT2022 dataset. Models such as CatBoost, Extreme Gradient Boost (XGBoost), and LightGBM were developed and evaluated. The dataset was normalized using the Normalizer and MinMaxScaler functions from the scikit-learn framework. Model training was conducted with an 80:20 fixed data split for training and testing, along with 5-fold cross-validation. Testing revealed that XGBoost with MinMaxScaler and the 80:20 split achieved the highest accuracy of 99.89%. However, accuracy decreased to 94.04% when using 5-fold cross-validation. Nevertheless, XGBoost with MinMaxScaler consistently demonstrated the fastest computation time across all schemes. For instance, it required only 15 seconds for the fixed split scheme compared to 59 seconds for 5-fold cross-validation. These findings highlight the efficiency and accuracy of XGBoost when combined with MinMaxScaler under specific validation schemes.]]>
Copyrights © 2024
