| Autor: |
Mathew, Rose Mary, Gunasundari, R., Lal, Sujesh P. |
| Předmět: |
|
| Zdroj: |
Journal of Soft Computing & Data Mining (JSCDM); 2024, Vol. 5 Issue 1, p15-30, 16p |
| Abstrakt: |
In various real-world domains, the problem of imbalanced data poses a significant challenge since it affects the efficiency and trustworthiness of machine learning models. This article investigates Explainable Artificial Intelligence (XAI) methods for studying models created on imbalanced datasets. The main objective of this paper is to assess models trained on DOSMOTE resampled balanced datasets. Using XAI techniques, the study seeks to understand better inner processes that lead to model decisions. The methodology involves combining DOSMOTE resampling with XAI to provide holistic evaluation through both qualitative and quantitative analysis. It should be noted that F1-Scores of balanced datasets improve significantly: from 76% to 87% for Web-Phishing; and from 58% to 73% for Hayes-Roth. This research highlights the need for XAI in enhancing interpretability of models trained on resampled imbalanced data sets. It also shows how resampling affects decision making in a model while performing and recommends investigating other resampling techniques or combinations with XAI methods aimed at improving model interpretability and transparency. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
