Editing and training for ALVOT, an evolutionary approach

In this paper, a new method, based on evolution strategies and genetic algorithms, for editing the sample data and training the supervised classification model ALVOT (voting algorithms), is proposed. Usually, this model is trained using Testor Theory, working with all available data. Nevertheless, in some problems, testors are not suitable because they can be too many to be useful. Additionally, in some situations, the classification stage must be done as fast as it is possible. ALVOT's classification time is proportional to the number of support sets and the number of sample objects. The proposed method allows finding an object subset with associated features' weights, which maximizes ALVOT classification quality, and with a support sets system of limited size. Some tests of the new method are exposed. Classification quality of the results is compared against typical testors option.