Decision Trees for Uncertain Data Based On Statistical Uniform Distribution

Abstract: Certain or classical decision trees are constructed for training data sets containing certain data. But in real life, in many cases, data is always uncertain. Hence many previous data mining techniques such as classification, clustering, regression and association rule mining etc. are inefficient or inaccurate or they must be reconsidered in managing uncertain data. Present study proposes an efficient and more accurate uncertain data management technique in data classification using decision trees. This new technique is modelled using uniform distribution and it is called Uniform Decision Trees for Uncertain Data (UDTUD). Uniform decision tree classifiers constructed for uncertain data are more accurate than Certain Decision Tree (CDT) classifiers constructed using certain data. There exists many models for uncertain data management but we propose Uniform distribution model for uncertain data management because it gives more accurate results for some training data sets. Applying data mining techniques to uncertain data is computationally costly. Extensive experiments have been conducted which show that classification accuracies obtained by UDTUD are more accurate than classification accuracies obtained by Certain Decision Trees (CDTs).

Keywords: Uniform distribution, uncertain data, certain data, decision tree, classification, data mining, machine learning