Abstract

    Logistic Regression Model (LRM) is considered as one of the most widely used statistical techniques for analyzing the relationship between variables with applications in different fields of research studies. Usually, the model seeks to predict the impact of one or several independent variables (predictor) on one dependent (categorical or response) variable through a probability function. When the LRM is constructed then the logistic parameters are estimated commonly using Maximum Likelihood Estimator (MLE), which maximizes the probability of obtaining the observed set of data. However, the maximization of the likelihood function is achieved using the most common iterative method called Newton Raphson Method (NRM); it is employed for solving the non-linear system of equations but it leads to serious numerical problems.

     In this thesis, the binary (Dichotomous) LRM is considered. In order to overcome the deficiency of using NRM then to suggest the parameter estimates, four different modifications for NRM namely, D-B; C-M-T; A-C-T; and L-W-W-Z, are proposed; each is an iterative method based on NRM. Moreover, the main algorithm for each approach is introduced. Hence, to identify the most efficient technique, based on the number of iterations, all these procedures are compared with each other and then with NRM, the L-W-W-Z method has a most efficient, because approach to a true value with less number iteration. Finally, a practical implementation for Binary LRM is given using data from a conducted survey.

 

 

posted:12-3-2017