Meaningful Error Estimations for Data Analysis

Abstract

Much work has been done in recent years on designing techniques used as support tools in the knowledge discovery process, particularly in classification tasks. In most cases it is assumed that the data where these techniques are applied is free of errors or the data was cleaned in a previous phase. However the data cleaning process represents a great amount of time and effort to the general knowledge discovery process.In this paper, we present preliminary results to devise a method to determine if the amount of errors in a dataset that will be processed by means of Naive Bayes classifier will influence the results. Our results may be used as a criterion to determine if it is necessary to carry out the data cleaning tasks over the data that will be processed by the classifier. Since the cleaning process takes a lot of time and effort our results are a helpful tool in the overall knowledge discovery process.