CLASSIFICATION OF ANEMIC PALESTINIAN CHILDREN USING THE MULTINOMIAL LOGISTIC REGRESSION MODEL.

Various estimation methods and optimization algorithms may be used to estimate the parameters of the multinomial logistic regression model. Most of them require justification of the assumptions about the underlying class densities. Therefore, failure to justify these assumptions may result in a great loss of performance. This paper aims to assess the performance of the main estimation methods and algorithms for building reliable multinomial logistic regression models. Seven estimation methods and algorithms are compared using different assessment techniques to arrive at a reliable multinomial logistic regression model for a given dataset. The result is that the ridge multinomial regression method proves to be the most reliable method with the highest area under the receiver operating characteristic (ROC), or ROC curve, and the lowest error rate for classifying children and identifying significant risk factors on anemia status among all other methods. A detailed description of the results of applying this method to a real dataset from a survey, conducted by the Palestinian Bureau of Statistics to classify children of less than five years of age (2010?2011) according to their anemia status, is illustrated. Ten independent variables from the survey are selected and used to classify children according to their anemia status (normal child, mild anemia, moderate anemia and severe anemia), a reliable multinomial regression model is built, and important risk factors of these anemia statuses are identified.