Discretization of Numerical Attributes in Rough Set Theory Based on Information Entropy with Heuristics Information

According to the consistency assumption in machine learning,the heuristics information of the data set statistic properties is used to select the discretization points from the candidate point set,in more detail,the mean and variance of data set are used to ascertain the region for searching optimal discretization points.A novel algorithm of numerical attributes discretization based on information entropy is proposed.The testing experiment with the UCI data sets has been performed.The results of experiment show that the discretization point set selected by using the new algorithm is the same as those by using the existing algorithm,and so does the results of decision tables discretization,but the time cost is different,the computing time of the new algorithm has saved about 40%~50% compared to the existing algorithm.