Network Anomaly Detection Using LightGBM: A Gradient Boosting Classifier

Anomaly detection systems are significant in recognizing intruders or suspicious activities by detecting unseen and unknown attacks. In this paper, we have worked on a benchmark network anomaly detection dataset UNSW-NB15, that reflects modern-day network traffic. Previous works on this dataset either lacked a proper validation approach or followed only one evaluation setup which made it difficult to compare their contributions with others using the same dataset but with different validation steps. In this paper, we have used a machine learning classifier LightGBM to perform binary classification on this dataset. We have presented a thorough study of the dataset with feature engineering, preprocessing, feature selection. We have evaluated the performance of our model using different experimental setups (used in several previous works) to clearly evaluate and compare with others. Using ten-fold cross-validation on the train, test, and combined (training and test) dataset, our model has achieved 97.21%, 98.33%, and 96.21% f1_scores, respectively. Also, the model fitted only on train data, achieved 92.96% f1_score on the separate test data. So our model also provides significant performance on unseen data. We have presented complete comparisons with the prior arts using all performance metrics available on them. And we have also shown that our model outperformed them in most metrics and thus can detect network anomalies better.