Identifying Most Predictive Items

Frequent itemsets and association rules are generally accepted concepts in analyzing item-based databases. The Apriori-framework was developed for analyzing categorical data. However, many data include numerical values. Therefore, most existing techniques transform numerical values to categorical values. The transformation is done such that the rules are optimal with respect to support or confidence. In this paper we choose a different approach for analyzing data with numerical and categorical data. We present methods to identify items, which have a strong impact on a given numerical attribute. With other words, we want to identify items, whose occurrence in an itemset allows us to make predictions about the distribution function of the numerical attribute.