A study of mining certain itemsets from uncertain data

Association rule mining is an important data analysis method for discovering associations within data. Recently, some researchers have extended association rule mining techniques to imprecise or uncertain data. However, the question arises as to how we can mine relevant and interesting patterns from uncertain data. Additionally, using the Sigma-count, the summation of a large number of itemsets with very small support may induce irrelevant associations. To this end, this study proposes a new approach to discover relevant patterns from uncertain data. This approach is based on the alpha-cut method allowing us to filter out the irrelevant patterns with small support. Furthermore, a correlation measure, also known as lift, is used to augment the support-confidence framework for association rules. Next, we develop an algorithm to discover relevant and interesting association rules from uncertain data. Experimental results from the survey data show that the proposed approach can help us to discover interesting and valuable patterns with high correlation.