Application of Discrimination and Classification on Diabetes Mellitus Data
Michael Asamoah-Boaheng

Abstract
The assignment/allocation of individuals/observations to the various known groups with known mean vectors and distinguishing characteristics has been a major concern for years and several attempts have been made at deriving parsimonious rules that address this hurdle. In this study, Fishers Linear Discriminant Function (FLDF) was derived to provide maximum separation between Type 2 and Type1diabetes patients based on identified risk factors. The assumptions of FLDF were achieved by BoxMtest of equality of covariance matrices. A seven variate data on 620 diabetes patients obtained from Komfo Anokye Teaching Hospital (KATH) diabetes ward was obtained and used for data analyses . The derived FLDF was used to reclassify the original observation to obtain the discriminant scores from the functions and 85.3 percent correct classification was achieved. Also 84.8 percent of the cross validated grouped cases were correctly classified into either being a Type 2 or Type 1 diabetes patient group. Patients age as well as their BMI were identified to be the two major contributing variables in classifying a patient as a type1 or type 2 diabetes.

Full Text: PDF