A Dataset Decomposition Approach to Data Mining and Machine Discovery

We present a novel data mining approach based on decomposition. In order to analyze a given dataset, the method decomposes it to a hierarchy of smaller and less complex datasets that can be analyzed independently. The method is experimentally evaluated on a real-world housing loans allocation dataset, showing that the decomposition can (1) discover meaningful intermediate concepts, (2) decompose a relatively complex dataset to datasets that are easy to analyze and comprehend, and (3) derive a classi er of high classi cation accuracy. We also show that human interaction has a positive e ect on both the comprehensibility and classi cation accuracy.

• Type: Conference
• Year: 1997
• Where: KDD
• Authors: Blaz Zupan, Marko Bohanec, Ivan Bratko, Bojan Cestnik
• Keywords: Classi Cation Accuracy, Complex Dataset, Data Mining, KDD 1997, Loans Allocation Dataset