FEATURE SELECTION FOR EFFECTIVE CALCULATION OF A SIMILARITY MEASURE

摘要

In any mining application for useful information from databases, an increasing number of features (attributes) makes worse results and loses much time. We propose a feature selection technique which saves computation time and does not spoil effect of mining. We take an algorithm called Iterated Contextual Distances (ICD), show its problems for practical applications, and propose a feature selection method, which mitigates these problems. Then we show effects of the feature selection by experiments performed on a real dataset.