Parameterless data compression and noise filtering using association rule mining

The explosion of raw data in our information age necessitates the use of unsupervised knowledge discovery techniques to understand mountains of data. Cluster analysis is suitable for this task because of its ability to discover natural groupings of objects without human intervention. However, noise in the data greatly affects clustering results. Existing clustering techniques use density-based, grid-based or resolution-based methods to handle noise but they require the fine-tuning of complex parameters. Moreover, for high-dimensional data that cannot be visualized by humans, this fine-tuning process is greatly impaired. There are several noise/outlier detection techniques but they too need suitable parameters. In this paper, we present a novel parameterless method of filtering noise using ideas borrowed from association rule mining. We term our technique, FLUID (Filtering Using Itemset Discovery). FLUID automatically discovers representative points in the dataset without any input parameter by mapping the dataset into a form suitable for frequent itemset discovery. After frequent itemsets are discovered, they are mapped back to their original form and become representative points of the original dataset. As such, FLUID accomplishes both data and noise reduction simultaneously, making it an ideal preprocessing step for cluster analysis. Experiments involving a prominent synthetic dataset prove the effectiveness and efficiency of FLUID.