Performance Analysis of Data Normalization Methods

Abstract

Statistical Data Normalization is a very important input preprocessing operation that should be done before data is fed into the training network. However, there is need for a suitable selection of normalization technique since normalization on the input has potential of varying the structure of the data and may impact on the outcome of the analysis. This paper investigates and evaluates some important statistical normalization techniques by studying thirty published papers that used wine dataset available in the UCI repository and their impact on performance accuracy. Results reveal that Min-Max normalization technique had the best performance accuracy of 95.91% on the average among all the other normalization types.