Feature Selection Using Typical Testors Applied to Estimation of Stellar Parameters

In this paper a comparative analysis of feature selection using typical testors applied on astronomical data, is presented. The comparison is based on the classification efficiency using typical testors as feature selection method against the classification efficiency using Ramirez (2001) method, which uses genetic algorithms. The well-known K-nearest neighbors rule (KNN) was used as classifier. The feature selection based on typical testors was modified to be applied on a prediction problem of a real valued function. The feature selection obtained with typical testors reduces the amount of features in approximately 50% and the classification error index is better than both using the original data and Ramirez's method.