Knowledge-Based Linguistic Attribute Hierarchy for Diabetes Diagnosis

A hierarchy of Linguistic Decision Trees (LDTs), called linguistic attribute hierarchy (LAH), can provide a transparent information propagation and a hierarchical decision making process. In this paper, we quantified the effect of various factors on the diagnosis of Diabetes with the information gain of each attribute to the decision variable, and developed an LAH, where, the LDTs are constructed under the framework of the knowledge-based label semantics, referring to the knowledge of the diagnosis criteria of Diabetes, defined by the World Health Organisation. A genetic wrapper algorithm was developed to find the best LAH for improving the accuracy of Diabetes diagnosis. The optimal LAH for Diabetes diagnosis achieved the accuracy up to 92% on the benchmark database, Pima Indian Diabetes data. The accuracy is much better than that in the research literature.