A Human-Centered Multiperspective and Interactive Visual Tool For Explainable Machine Learning

Abstract

Listen

Translate article

Understanding why a trained machine learning model makes some decisions is paramount to trusting the model and applying its recommendations in real-world applications. In this article, we present the design and development of an interactive and visual approach to support the use, interpretation and refinement of ML models, whose development was guided by user's needs. We also present Explain-ML, an interactive tool that implements a visual multi-perspective approach to the support interpretation of ML models. Explain-ML development followed a Human-Centered Machine Learning strategy guided by the target (knowledgeable) users' demands, resulting in a multi-perspective approach in which interpretability is supported by a set of complementary visualizations under several perspectives (e.g., global and local). We performed a qualitative evaluation of the tool´s approach to interpretation with a group of target users, focused on their perspective regarding Explain-ML helpfulness and usefulness in comprehending the outcomes of ML models. The evaluation also explored users' capability in applying the knowledge obtained from the tool's explanations for adapting/improving the current models. Results show that Explain-ML provides a broad account of the model's execution (including historical), offering users an ample and flexible exploration space to make different decisions and conduct distinct analyses. Users stated the tool was very useful and that they would be interested in using it in their daily activities.