Mining Association Rules from Data with Missing Values by Database Partitioning and Merging

Often, real world applications contain many missing values. In mining association rules from real datasets, treating missing values is an important problem. In this paper, we propose a pattern-growth based algorithm for mining association rules from data with missing values. No data imputations are performed. Each association rule is evaluated using all the data records with which attributes of it are not missing values. Our algorithm partitions the database so that the data record with which the same attributes contain missing values is assigned to the same database partition, and the algorithm mines association rules by combining these database partitions. We propose methods of reducing processing workload: estimating the upper bound of global support using local supports, reutilizing part of the constructed tree structure, and merging redundant database partitions. Our performance study shows that our algorithm is efficient and can always find all association rules