Uncovering Population Trends with Actionable Clustering

Clustering is a widely-used data mining tool, which aims to discover groups of similar items in data. We introduce a new clustering paradigm, actionable clustering, that enables the discovery of population trends. Unlike previous clustering paradigms that aim to understand relationships amongst the individual members, the goal of actionable clustering is to uncover trends at the populations level (e.g. treatment groups) through the analysis of their members (e.g. subjects across different treatments). Insights gained from the trends of populations can often support a call to action that cannot be informed through previous clustering methods. We propose the first actionable clustering algorithm, and prove that it finds near-optimal solutions when data possesses inherent cluster structure. The insights revealed by actionable clusterings enabled experts in the field of medicine to isolate successful treatments for a neurodegenerative disease, and those in finance to discover patterns of unnecessary spending. We also perform experiments on six UCI datasets, which show that our algorithm outputs a superior quality actionable clustering when compared with its adapted competitors.