Mining Frequent Itemsets from Noisy Data

Abstract

As we face huge amounts of varied information, data mining, which helps us discover hidden features or rules from voluminous data systematically, has become more important [3, 4, 6, 10]. However, real world data is often dirty, including noise such as missing or irrelevant values. The information mined from such noisy data may be incorrect. We model noisy data with probabilities, assuming that noise is mixed with data statistically. We also propose a way to find frequent itemsets [2] by estimating supports on noiseless data from noisy data. An algorithm using FP-tree [6, 10] is also presented to mine frequent itemsets efficiently.