Classification models comparison from the user’s level of interpretability
Lenka Dobranská, Anna Biceková, František Babič · 2023
The effective deployment of classification models generated with the help of relevant machine learning algorithms also affects the ability of target users to understand the given models in the context of the solved task, especially in the domain of healthcare provides or medical diagnostics. Various interpretation methods and interpretability assessment metrics have been proposed to provide insight into the decision-making processes of these models. Therefore, we decided to propose an approach that, based on the user’s preferences entered at the input, will evaluate individual models and sort them according to the preferences of traditional or interpretability metrics. Our approach uses the TOPSIS method for evaluation and the Saaty’s matric to represent the user’s preferences. The first testing with public available Breast Cancer Wisconsin dataset brough plausible results for the future work.