Assessing the Models with Resampled Data Using Explainable Artificial Intelligence Techniques

Abstract

In various real-world domains, data imbalance poses a prevalent challenge, significantly affecting the effectiveness and reliability of machine learning models. This paper addresses this issue by exploring the application of Explainable Artificial Intelligence (XAI) techniques to gain insights into machine learning models developed using imbalanced datasets with multiple classes.  The primary objective of this article is to evaluate machine learning models trained on imbalanced datasets that have undergone the DOSMOTE resampling procedure to achieve balance. The study aims to utilize XAI methods to gain a deeper understanding of the internal processes and explanations underlying the decisions made by these models. The methodology employed in this article involves a combination of resampling the imbalanced training data using DOSMOTE and applying XAI techniques for model assessment. This method combines both quantitative analyses, involving the application of machine learning models to the resampled data, and qualitative analysis, which focuses on utilizing XAI techniques to interpret and comprehend the decisions of the models. The specific results and outcomes of the study are presented within the relevant sections of the article. The study assessed the performance of models trained on resampled datasets and gained insights into their decision-making processes through XAI techniques. In conclusion, this article highlights the significance of providing better interpretability and understanding of machine learning models trained on resampled, imbalanced datasets using XAI techniques. The study sheds light on the impact of the resampling procedure on model performance and how the models make decisions following training on balanced data. Future research could explore alternative resampling techniques or the combination of multiple XAI methods to further enhance the interpretability and transparency of machine learning models trained on imbalanced data.