A rapid cross-validation computing for three-way decisions in imbalanced data

Abstract

Three-way decisions (TWDs) developed from rough set theory play a crucial role in decision-making and have been widely applied across various scenarios. However, the prevalence of imbalanced data in real-world applications poses significant challenges to TWDs. Traditional TWD approaches often overlook the impact of imbalanced data, leading to suboptimal performance when applied to datasets with non-uniform class distributions. Stratified K-fold cross-validation is a popular technique for evaluating models on imbalanced datasets. In this paper, we introduce stratified K-fold based cross-validation to TWDs, so as to enhance the models' reliability and accuracy. Nonetheless, implementing stratified K-fold cross-validation to TWDs requires training the models K-times, leading to high computational complexity. By analyzing the data models for stratified K-fold cross-validation, we provide an approach of carrying out rapid validation in TWDs via reducing computation as much as possible, so as to improve the overall performance. Theoretical results can guarantee the correctness of the provided techniques. We conduct experiments on widely-used machine learning datasets. The experiment results demonstrate that the proposed method significantly enhances computational efficiency while preserving model evaluation accuracy and offering strong stability for TWD thresholds. This paper provides a validation tool and reasoning method for dealing with imbalanced data in TWD.