Contrast Set Mining Through Subgroup Discovery Applied to Brain Ischaemina Data

Abstract

Abstract. Contrast set mining aims at finding differences between different groups. This paper shows that a contrast set mining task can be transformed to a subgroup discovery task whose goal is to find descriptions of groups of individuals with unusual distributional characteristics with respect to the given property of interest. The proposed approach to contrast set mining through subgroup discovery was successfully applied to the analysis of records of patients with brain stroke (confirmed by a positive CT test), in contrast with patients with other neurological symptoms and disorders (having normal CT test results). Detection of coexisting risk factors, as well as description of characteristic patient subpopulations are important outcomes of the analysis.

Contrast Set Mining for Distinguishing Between Similar Diseases

“…As shown in [7], the subgroup discovery approach to contrast set mining has several advantages. Its application in brain ischemia data analysis has shown that sometimes the right task to address is one-vs-all contrast set mining rather then the classical round robin formulation of contrast set mining.…”

“…Even though the definitions of subgroup discovery and contrast set mining seem different, the tasks are compatible [7]. From a dataset of class labeled instances (the class label being the property of interest) by means of subgroup discovery [4] we can find contrast sets in a form of short interpretable rules.…”

“…Contrast set mining can be performed by a specialized algorithm STUCCO [1], through decision tree induction and rule learning [13], and-as shown in our recent work-through subgroup discovery [7]. Section 3 presents the results of decision tree induction on our contrast set mining task and discuss advantages and disadvantages of this approach.…”

“…Section 3 presents the results of decision tree induction on our contrast set mining task and discuss advantages and disadvantages of this approach. In Section 4 we show an approach to contrast set mining through subgroup discovery by providing a mathematically correct translation from contrast set mining to subgroup discovery [7] and an implementation of the approach in the Orange [2] open source data mining toolbox. Next we show that the direct "round robin" contrast set mining approach to solve our descriptive induction task leads to rather disappointing results.…”

Tight Optimistic Estimates for Fast Subgroup Discovery

“…It is a general approach that has shown to be useful in a variety of application scenarios, like medical consultation systems [ABP06], spatial analysis [KM02], marketing campaign planning [LCGF04], and also in contrast set mining tasks [KLGK07].…”