Predicting the risk of gestational diabetes mellitus through nearest neighbor classification

Abstract

Gestational diabetes mellitus (GDM) may arise as a complication of pregnancy and can adversely affect both mother and child. Diagnosis of this condition is carried out through screening coupled with an oral glucose test. This procedure is costly and time-consuming. Therefore, it would be desirable if a clinical risk assessment method could filter out any individuals who are not at risk of acquiring this disease. This problem can be tackled as a binary classification problem. In this study, our aim is to compare and contrast the results obtained through binary logistic regression (BLR), used in previous studies, and three well-known non-parametric classification techniques, namely the k-nearest neighbors (kNN) method, the fixed-radius-NN method and the kernel-NN method. These techniques were selected due to their relative simplicity, applicability, lack of assumptions and nice theoretical properties. The test dataset contains information related to 1,368 subjects across 11 Mediterranean countries. Using various performance measures, the results revealed that NN methods succeeded in outperforming the BLR method.