Mid-infrared spectroscopy with an effective variable selection method based on MPA for glucose detection

Mid-infrared absorption spectroscopy is one of the most promising non-invasive blood glucose detection tech-nologies. However, subsistent blood glucose detectors based on mid-infrared absorption spectroscopy technology are incapable of meeting the requirement of superior accuracy. Aimed at improving the prediction performance of glucose concentration in mid-infrared (MIR) spectra, an effective variable selection method (EVS) is proposed that generates the importance index for wavenumber to reduce the time cost. Tournament selection is then applied to avoid the reduction of exploration ability. L1 regularization is finally performed to tackle overfitting. 66 variables have been retained in the glucose fingerprint spectra by the proposed method, and the corre-sponding R and RMSEP are 0.93 and 32.22 mg/dL, respectively. Experimental results show that the proposed method outperforms five alternative approaches for variable selection from relevant literature, which is more accurate and robust.

Automated summary

Mid-infrared absorption spectroscopy is a promising technique for non-invasive blood glucose detection. However, current detectors based on this technology lack the required accuracy. In order to improve the prediction performance, an effective variable selection method is proposed, which generates importance index for wavenumber to reduce time cost, applies tournament selection to maintain exploration ability, and uses L1 regularization to address overfitting. Experimental results demonstrate that this method outperforms alternative approaches and achieves higher accuracy and robustness in variable selection.