Mining fuzzy association rules from heterogeneous probabilistic datasets

Association rule mining (ARM), as a useful method to discover relations between attributes of objects, has been widely studied. The previous methods focused on ARM either from a certain dataset with different type attributes, or from a probabilistic dataset with only Boolean attributes. However, little work on ARM from a probabilistic dataset with coexistence of different type attributes has been mentioned. Such dataset is named Heterogeneous Probabilistic Dataset (HPD), which is prevalent in the real-world applications. This paper develops a generic framework to discover association rules from a HPD. Considering the different type data in the dataset, we first convert a HPD to a probabilistic dataset with fuzzy sets by fuzzification. A novel Shannon-like Entropy is then introduced to measure the information of an item with coexistence of fuzzy uncertainty hidden in different type data and random uncertainty in the transformed dataset. Based on this Shannon-like Entropy, Support and Confidence degrees for such multi-uncertain dataset are defined. Finally, we design an Apriori-like algorithm to mine association rules from a HPD using the above measures. Experimental results show that the proposed algorithm for HPD is feasible and effective.