Ml and Statistics for Trend Prognosis of Complaints in the Automobile Industry

This paper describes the application of clustering as preprocessing step and two diierent KDD-algorithms in a real world application taken from the automobile industry. We analyze the ability of these algorithms to get a trend prognosis of fault rate behavior. Furthermore we show the problem that at the beginning of the KDD process data in the range of megabytes is available, but that this is not enough for our special purposes and how we used hierarchical clustering in the preprocess-ing to get a solution for this problem.