Using Multinomial Logistic Regression and Artificial Neural Networks in the Classification of Living Standard in Palestine

Abstract:

The world today is encountering many global issues: political, social and economic. Whereas, improving the standard of living and reducing income inequality; especially, in Palestine are the basic concerns to be solved nowadays.

In this study, Multinomial Logistic Regression (MLR) and Artificial Neural Network (ANN) were discussed on Expenditure and Consumption Survey Data in 2011. The two methods were compared according to accuracy assessment, ROC curve,  AIC and BIC assessment criterion.

The study contained two  parts, theoretical part and application part in which data was used from Expenditure and Consumption Survey (2011).Data consisted of (4317) households from the West Bank and Gaza Strip. Data contained (12) variables, where the dependent variable was the standard of living, which was ordinal variable and contained of three categories ( High, Middle and low standard of living ) and (11) other independent variables. 

The study aims to choose the best statistical model for Palestinian standard of living data. Two models for each method were tested by group of statistical tests to define the best model. The results showed that Artificial Neural Network method was better than Multinomial Logistic Regression method, where Artificial Neural Network ratio reached 90.1% compared to Multinomial Logistic Regression ratio, which reached 77.7%.Area under ROC curve for Artificial Neural Network analysis reached 97.2%,while (AUC) area under curve analysis model reached 89.9% for Multinomial Logistic Regression. While comparing was done using AIC and BIC  assessment criterion.