Evaluation of Class Binarization and Feature Selection in Tear Film Classification using TOPSIS

Dry eye syndrome is a prevalent disease which affects a wide range of the population and can be diagnosed through an automatic technique for tear film lipid layer classification. In this setting, class binarization techniques and feature selection are powerful methods to reduce the size of the output and input spaces, respectively. These approaches are expected to reduce the complexity of the multi-class problem of tear film classification. In previous researches, several machine learning algorithms have been tried and only evaluated in terms of accuracy. Up to now, the evaluation of artificial neural networks (ANNs) has not been done in depth. This paper presents a methodology to evaluate the classification performance of ANNs using several measures. For this purpose, the multiple-criteria decision-making method called TOPSIS has been used. The results obtained demonstrate that class binarization and feature selection improves the performance of ANNs on tear film classification.