USING MULTINOMIAL LOGISTIC REGRESSION ANALYSIS IN ARTIFICIAL NEURAL NETWORK: AN APPLICATION

Determination of Artificial Neural Networks' classification and parameter estimation with Multinomial Logistic Regression Analysis was examined in this study. One of the modeling types suggested in case of having the dependent variables in categorized/classified structure and the independent variables in different structures such as nominal, ordinal, and intervals etc. in a research pattern is "Multinomial Logistic Regression (MLR)" method. MLR and Artificial Neural Networks (ANN) based MLR Analyses' findings were studied comparatively in the model, where the dependent variable performed categorical structure and the independent variables performed mixed (continuous-discrete) structure. For the research, real data that were gathered in the context of the study entitled "Studying Primary School Students' Views on their Communications with the Teachers and the Expected Situation" were used by the "Students' Expectations from their Teachers in Teacher-Student Communication Process Scale" developed by Dogan (2009). Within the context of this study, the total score obtained from the scale was assigned as the dependent variable, and variables such as the school type (public-private), gender, grade, mother's profession, father's profession, mother's educational status, father's educational status, number of brothers/sisters, monthly income, and internet usage time were assigned as independent variables. ANN has classified the dependent variable in high correctness level and showed the model's fit in a higher level than MLR. Moreover, ANN has obtained parameter coefficients unlike MLR. It was considered that the model studied was estimated more consistently and correctly with ANN.