The explainability challenge in descriptive analytics: do we understand the data?

Description

Nowadays, ever more data is collected, for instance in the healthcare domain. The amount of patients’ data has doubled in the previous two years. This exponential growth creates a data flood that is hard to handle by decision makers. In many domains, humans are collaborating with machines for decision making purposes to cope with the resulting data complexity and size. This collaboration can be realized through machine learning, visual analytics, or online analytical processing, where a machine is just a tool – but often used to take important decisions. The question now is: do we really understand the data using the tool this way?
Explainability is a great challenge in data analytics, with the aim to explain to the user why certain decisions have been recommended or made. This challenge is especially important in predictive and prescriptive analytics. Less attention with this respect is payed to less mature analytics levels, descriptive and diagnostics, although they are the first steps for understanding data.
Data analysis methods use numbers, figures or mathematical equations to show data, decision recommendations and patterns. Yet for a human, the natural way of communication is natural language: words, not numbers or figures. This causes a gap between the meaning of data and human understanding. The challenge is: How to make data more understandable for humans?
Fuzzy techniques, or the application of the computing with words paradigm has the potential to close the gap by using natural language as the communication means. In this talk I focus on descriptive analytics and show with a set of examples how fuzzy techniques can provide better insight of data to the user. I pay special attention to the technique of linguistic summaries.

Period	27 Feb 2019
Held at	IBM, United States, California