A Diabetes Risk Assessment Model Using Limited Vitro Physiological Indicators

Abstract

A majority of reported diabetes assessment models using physiological indicators as input features use vitro physiological indicators such as blood glucose, insulin, which are difficult to measure and unobtained in daily life, leading to low practicality of reported models. To overcome this issue, this paper proposed a diabetes risk assessment model using limited vitro physiological indicators, which can be measured and obtained easily. This model provided advice about diabetes risk based on the values of input physiological indicators, in this way alerting people at high diabetes risk to prevent diabetes in early stages. Feature-weighted k-Nearest Neighbors (FW-kNN) algorithm was used to build the model, manta ray foraging optimization (MRFO) algorithm was used for searching the optimal feature weights, 10-fold cross-validation and 4:1 training-testing method were adapted to evaluate the performance of the proposed FW-kNN algorithm. The results of experiments revealed that the proposed FW-kNN algorithm achieved 70.9% accuracy using 10-fold cross-validation and the area under receiver operating characteristic (ROC) curve (AUC) was 0.80, which proved the proposed FW-kNN algorithm had a good performance and the outputs of the proposed FW- kNN model were of reference to diabetes risk assessment.