Feature Selection for Optimal Weather Detection with Meteorological Radar Data

In this paper, the data of Dual-Pole Meteorology Radar which is taken from the MGM and located in Hatay, is used for developed weather detection with feature selection. Classical classification methods were used first to weather detection. After the results of this classification, studies were conducted to select only the features that are critical to classification, rather than all features, with a view to reducing the level of performance with fewer features. These studies were conducted on two classifiers; a classifier was first used to detect "rainfall" or "no rainfall" in the region and then classified for "bird-insect" or "clutter". From the total of eight features found in our database, the most important and most useful features for both classifiers have been determined. The results show that the feature selection method we have developed has a similar performance when a few attributes are used instead of all. Thus, it is possible to achieve similar classification performance with lower calculation capacity.