Attribute weighting with probability estimation trees for improving probability-based ranking in liver diagnosis

In traditional medical diagnosis, the aim of using classification learning models is for high diagnosis accuracy. However, accurate class probability estimate is more desirable than prediction accuracy in practical medical diagnosis application. Although several probability estimation models based on decision trees have been adopted in many other areas, they both have an obstacle to achieving accurate probability prediction, for example time-consuming and equal weighted input. In addition, the research and application of probability estimation trees models in traditional Chinese medicine diagnosis area are still very insufficient. So in this paper, we propose our method to overcome these problems, and compare our method with several representative methods, for example traditional decision trees, C4.4, NBTree, CITree and CLLTree, measured by classification accuracy, AUC and Conditional Log Likelihood (CLL) on Cirrhosis and Hepatitis traditional Chinese medicine sample sets. From our experiments, the proposed algorithm can efficiently improve the performance of models and yield more accurate probability prediction than those representative models. Our proposed method performs well in the field of TCM diagnosis.