A Multi-Feature Quantile Regression Approach for Insulin Sensitivity Prediction

A multi-feature quantile regression method is presented in this work to forecast insulin sensitivity of critical care patients. The suggested approach delivers a 90 % confidence interval for future insulin sensitivity by combining blood glucose measurements with insulin sensitivity values. Predicting insulin sensitivity distributions within 5th and 95th percentiles is achieved using neural network combined with quantile regression. Adapted model can scored narrower prediction confidence interval. However, the empirical converge within 90 %, thus, improving the prediction efficacy. Comparing the results with the conventional one input feature model, the proposed method results perform narrower prediction confidence interval. These findings imply multi-feature method can improve glycemic control techniques and assist in making more exact insulin dosage which is clinically significance in critical care environments. Including BG data results in a complete depiction of patient metabolic dynamics, according