The performance comparison of models for predicting the risk of losing student loan by fuzzy neural network method and multiple linear regression analysis method
This research presents the comparing the performance of the prediction model of the risk of losing student loan by Fuzzy Neural Network (FNN) and Multiple Linear Regression Analysis (MLR) . We used the preparation of 296 samples, assigned the income variables, which are characterised by Fuzzy attribute and had problem-solving with Dummy variables. Also, we proposed the method of converting fuzzy data by making a fuzzy attribute matching of Neural Network (NN) and Multiple Linear Regression. It calculates the fuzzy linguistic term, the fuzzy language and the Crisp value, and after that have filtered variables with Pearson correlation technique and multiple linear regression which get eight independent variables and RiskForPay is a dependent variable. Results from this research, the appropriate model of Fuzzy Neural Network is a division of learning data, and cross-validation fold is 5 with an accuracy of 83.33% +/- 6.02% which models have 8-5-6 structure, momentum 0.2 and learning rate 0.3. The predictive model with multiple linear regression equations has Root Mean Squared Error: 1.513 +/- 0.000 and Squared Correlation: 0.081, respectively.