Noninvasive Blood Glucose Detection Using an Improved Sparrow Search Algorithm Combined with an Extreme Learning Machine Based on Near-Infrared Spectroscopy

The traditional way of measuring blood glucose causes pain and inconvenience to patients. Near-infrared spectroscopy is a promising noninvasive alternative. However, the prediction accuracy of the currently used quantitative blood glucose model for near-infrared spectroscopy decreases when a patient's physiological state changes. Therefore, we propose an improved sparrow search algorithm (ISSA) to optimize the initial weights and thresholds of extreme learning machines (ELM) in this paper. We used a tent chaotic map to improve the diversity of the SSA population. We also adopted reverse learning to initialize the population and expand the population search range, which further improved the search performance of the SSA. The predicted results of the ISSA–ELM model were more accurate and generalizable than those of the SSA–ELM model. Clarke error grid analysis showed that the proportion of predicted samples falling into the A region was 90%, and the proportion falling into the B area was 10%, which is in accordance with clinical requirements. Therefore, this model has strong potential for application in non-invasive detection of human blood glucose.