Artificial Neural Network Model Based Estimation of Finite Population Total

Robert Kasisi, Romanus O. Odhiambo, Anthony G. Waititu

Abstract: In a finite population denoted U=1, 2, 3,. . . . . , N of N identifiable units, let X be the survey variable. The survey variables are observations from a super population. It is possible to get total information about these survey variables such as their total population, mean or their variance. In most cases auxiliary information about X is provided. A simple approach to using this auxiliary information is to assume a working model that describes the connection between the study variable of interest and the auxiliary variables. Estimators are then derived on the basis of this model. The best estimators are the ones that have good efficiency if the model is true, and are consistent if the model is inappropriate. In this study, we derive a nonparametric artificial neural network estimator of finite population total. The estimator is design unbiased, design consistent and asymptotic normal.

Keywords: superpopulation, auxiliary information, artificial neural networks, sample, survey sampling