Adjusting Fuzzy Similarity Functions for use with standard data mining tools

作者：Meged Avichai; Gelbard Roy^*

来源：Journal of Systems and Software, 2011, 84(12): 2374-2383.

DOI：10.1016/j.jss.2011.07.009

• Data representation
• Data mining
• Data segmentation
• Classification
• Clustering
• Similarity function
• Similarity measure

Data mining is crucial in many areas and there are ongoing efforts to improve its effectiveness in both the scientific and the business world. There is an obvious need to improve the outcomes of mining techniques such as clustering and other classifiers without abandoning the standard mining tools that are popular with researchers and practitioners alike. Currently, however, standard tools do not have the flexibility to control similarity relations between attribute values, a critical feature in improving mining-clustering results. The study presented here introduces the Similarity Adjustment Model (SAM) where adjusted Fuzzy Similarity Functions (FSF) control similarity relations between attribute values and hence ameliorate clustering results obtained with standard data mining tools such as SPSS and SAS. The SAM draws on principles of binary database representation models and employs FSF adjusted via an iterative learning process that yields improved segmentation regardless of the choice of mining-clustering algorithm. The SAM model is illustrated and evaluated on three common datasets with the standard SPSS package. The datasets were run with several clustering algorithms. Comparison of "Naive" runs (which used original data) and "Fuzzy" runs (which used SAM) shows that the SAM improves segmentation in all cases.

• 出版日期2011-12