IREX: Iterative Refinement and Explanation of classification models for tabular datasets

Abstract

Tabular datasets, collections of rows and columns, are fundamental in data analysis in basically all areas of research. Self-report questionnaires are a very common and useful tool for gathering data from users, patients, or customers. Often, experts can label each item of these questionnaires as a measure of a given condition or behavior, e.g., mental health conditions such as depression. Considering this, many artificial intelligence techniques, particularly those related to machine learning, have been proposed to analyze the data provided by these tools. However, self-report questionnaires can be very extensive, which often affects the quality of the responses, complicates the data analysis, and renders them more time-consuming. In this paper, the software IREX is presented. IREX iteratively refines tabular datasets, such as self-report questionnaires, while providing an explanation of a given classification model.

Keywords

• Explainable Artificial Intelligence
• ML model refinement
• Tabular data
• Depression screening