OPTIMALITY CRITERIA FOR CLASSIFICATION THROUGH ROC CURVE ANALYSIS IN THE PRESENCE OF OUTLIERS

In this paper, we focus on fitting non-parametric a nd Binormal ROC Curves using spreadsheet functions and illustrate the procedures by using TB data coll ected in a hospital study. The Receiver Operating Characteristic (ROC) curves is one of those statist ical techniques that are now ubiquitous in a wide variety of substantive fields and it is a well acce pted measure of accuracy for tests with both contin uous and ordinal results. Tuberculosis (TB) is a major h ealth problem often called “the captain of these me n of death”. Pleural Fluid Adenosine Deaminase (ADA) is a known biomarker to determine the presence to TB as against the gold standard called IFN-Gamma which is both costly and not affordable to many patients. The optimal cutoff value which provides a better accuracy of classification is determined. The effect of age and sex as cofactors is also consider ed for classification of patients using linear disc riminant function (LDF). In this paper, we have studied the discriminating ability of pleural fluid ADA in the diagnosis of TB. The ROC analysis has shown that a cut-off value of 36 IU/L provides the optimal trade off between sensitivity and specificity. It is obse rved that the age and sex of the subject may influe nce the decision criterion.