An intelligent assistant for exploratory data analysis

In this paper we present an account of the main features of SNOUT, an intelligent assistant for exploratory data analysis (EDA) of social science survey data that incorporates a range of data mining techniques. EDA has much in common with existing data mining techniques: its main objective is to help an investigator reach an understanding of the important relationships ina data set rather than simply develop predictive models for selectd variables. Brief descriptions of a number of novel techniques developed for use in SNOUT are presented. These include heuristic variable level inference and classification, automatic category formation, the use of similarity trees to identify groups of related variables, interactive decision tree construction and model selection using a genetic algorithm.