Discretization algorithms play an important role in many areas such as data mining, machine learning and artificial intelligence. Therefore, a heuristic discretization technique based on the class-attribute interdependence is proposed. A new discretization criterion is defined, which selects best cut points in terms of characteristics of the data itself and overcomes the existing deficiencies of state-of-the-art top-down discretization methods. A novel measure of inconsistency based on variable precision rough sets(VPRS) model is developed, which effectively controls information loss generated by discretization and allows an adaptive degree of misclassification. Empirical experiments and statistical analysis show that the proposed technique generates a better discretization scheme which significantly improves the accuracy of classification by running J4.8 and SVM.