Error-based and entropy-based discretization of continuous features

摘要: We present a comparison of error-based and entropy-based methods for discretization continuous features. Our study includes both an extensive empirical as well analysis scenarios where error minimization may be inappropriate criterion. method based on the C4.5 decision tree algorithm compare it to existing algorithm, which employs Minimum Description Length Principle, recently proposed technique. evaluate these with respect Naive-Bayesian classifiers datasets from UCI repository analyze computational complexity each method. results indicate that MDL heuristic outperforms average. then shortcomings approaches in methods.