Enhancing Classifiers Performance using Data Quality Checking

Classification is an important data mining. Practically, datasets containing both redundancy and conflicting data have to be dealt with. Their presence degrades the performance of any classification algorithm. Besides performing 'Data Tuning', this paper aims to solve the exceptions that occur in classifiers during rule generation and identification of class labels. It suggests the EUTOME procedure to improve the performance of classifiers during both training and testing procedures. A far-flung experimental evaluation on a number of real and synthetic databases shows that EUTOME is state-of-the-art classification algorithm. Exceptions occur when the prediction of a class label cannot be determined by majority voting. The influence factor that has been calculated for each attribute, for exception handling.