Trusting Classifiers with Interpretable Machine Learning Based Feature Selection Backpropagation

In a machine learning classification problem, feature selection is a required pre-processing phase which identifies important and relevant features from the dataset to potentially reduce the computational complexity and improves the overall classification performances. Feature reduction mechanisms, such as Information Gain, Gain Ratio, Chi-squared, ReliefF, Deep Learning, etc. along with domain knowledge are used to find the appropriate features from a dataset. In this paper, we propose a novel feature selection process based on interpretable machine learning technique (IMLFS) to find the optimal relevant features in detecting DDoS cyber-attacks. Based on the effectiveness of critical features, this technique is also used to explain a detected DDoS attack. These relevant features are used in the feature selection phase to retrain the model for better accuracy. The benchmark dataset, NSL-KDD is used to evaluate the proposed approach. Moreover, using the extracted features obtained from this dataset, we investigated our recently developed ensemble supervised framework. This investigation confirms the efficacy of the IMLFS approach by producing both higher detection accuracy and lower false positive alarms. A significant improved accuracy and model training times compared to earlier studies that compared various IML methods are reported here.