A Picture is Worth a Thousand Words: Visual Model Evaluation in Data Science Applications

Besides programming and mathematical or statistical skills, domain knowledge about the investigated problem is an important factor in data science. In practical applications, however, there often exist a gap between data scientists who have the technical skills in advanced analytics methods and the domain experts like managers and decision makers with substantial knowledge in their field. Less participation of domain experts during the complex model building and evaluation process of the data science pipeline can lead to acceptance problems and prejudices against the results. Moreover, data scientists depend on the domain knowledge of experts when it comes to evaluation and problem identification of their models. In this paper we address this issue by introducing an easy-to-understand heat map visualization technique for model evaluation. It enables all parties to discuss the analysis on the same level of complexity. The benefits are demonstrated based on a real world business example.