Abstract—Diabetes mellitus is a chronic disease that reduces quality of life since it often causes other complications such as heart disease, stroke, high blood pressure, liver disease, kidney disease, neuropathy and the loss of some organs in the body. This work proposes a temporal features extraction model which extracts the features embedded in historical data such as health examination data for classification. The proposed model can be used with any promising classification methods such as Naïve Bayes, Logistic Regression, C4.5 (J48), Bagging and SVMs. The extended temporal features can improve the accuracy and F-measure of the classification. This work evaluates the proposed method on health examination data during 2004-2010 (7 years) of factory employees in Thailand. It consists of 43,523 employees in total where 28,808 employees have only one record and 14,715 employees is examined more than once. Features used for diabetes classification are categorized into three groups: Physical Examination,Urinalysis and Biochemistry. The experiments show that data with temporal features gives the 97.25% accuracy and 0.57 F-measure which is a lot higher than data without temporal features.

Index Terms—Temporal model, classification, diabetes, datamining, healthcare.

Authors are with the Faculty of Information Technology, Rangsit University, Pathumthani, Thailand (e-mail: s.kritsada@it.rsu.ac.th, tel.: +66-2-9972200 ext. 4068; fax: +66-2-9972200 ext.4076).

Cite: Kritsada Sriphaew, Somphop Pathomnop, and M. L. Kulthon Kasemsan, "Temporal Data Classification of Diabetes Mellitus on Health Examination Data of Factory Employees," International Journal of Computer and Communication Engineering vol. 1, no. 1, pp. 31-34 , 2012.