Credit risk is a risk due to the failure or inability of the customer to return the amount of credit obtained from the company and its interest according to a predetermined or scheduled period of time. The main task of the credit risk classification method is to provide a separation between those who have the potential to fail and those who have not failed in terms of credit payments. The k-Nearest Neighbor (kNN) method as the most popular, simple and easily implemented machine learning method can be used to classify credit risk. However, its success depends on the number of neighbors or neighbors (k) applied and the relationship between each data with a class is rigid (crisp) where each data only has a relationship with one class exclusively, while the other classes have no relationship at all. This study proposes the incorporation of the principles of fuzzy logic into k-NN to minimize the stiffness that results in a new method known as Fuzzy k-Nearest Neighbor or Fk-NN. However, the fuzzy strength factor (m) and the number of neighbors (k) as the fundamental determinants of Fk-NN which have a direct impact on the accuracy generated by the model, the determination is often not easy and difficult to control, so the Modified method is proposed Particle Swarm Optimization (MPSO) to be able to help Fk-NN find the best m and k values non-manually. The results of the classification of credit risk data are 1000 data, with 900 composition of training data (90%) and 100 data (10%) of test data using Fk-NN with MPSO producing accuracy reaching 92.4%, with the best k value is 7 and the best m value is 9.
Copyrights © 2020