OBJECTIVES. The objectives of this paper are 1) to propose new techniques to learn and improve the multi-criteria decision analysis (MCDA) method PROAFTN based on machine learning approaches, and 2) to compare the performance of the developed methods with other well-known machine learning classification algorithms. METH-ODS. The proposed learning methods consist of two stages: the first stage involves using the discretization techniques to obtain the required parameters for the PROAFTN method, and the second stage is the development of a new inductive approach to construct PROAFTN prototypes for classification. RESULTS. The comparative study is based on the generated classification accuracy of the algorithms on the datasets. For further robust analysis of the experiments, we used the Friedman statistical measure with the corresponding posthoc tests. CONCLUSION. The proposed approaches significantly improved the performance of the classification method PROAFTN. Based on the generated results on the same datasets, PROAFTN 1 outperforms widely used classification algorithms. Furthermore, the method is simple, no preprocessing is required, and no loss of information during learning.