The Effect of Training Set Selection in Meteorological Data Mining

Abstract

We use meteorological data from the European Centre for Medium-Range Weather Forecasts (ECMWF) and the Meteorological Station of Mikra (Thessaloniki, Greece) as input to five data mining algorithms with the aim to build classification models for the prediction of the occurrence of precipitation in the station. We focus our study on the effect the selection of the training set has on the performance of the algorithms and more specifically, we attempt to determine the minimum training set size that can ensure effective application of the data mining techniques.