Interpretable Machine Learning Regressors for Mild Hypothermia Prediction in General Surgical Operations

Abstract

Hypothermia is a medical emergency that occurs when there is a low body temperature from the normal body temperature of 35oC. The occurrence of this emergency reportedly ranges from 33% to 89% during general surgical operations and often leads to extremely short and long-term complications. Fortunately, there has been a growing trend in using electronics and informatics for smart healthcare, particularly in using artificial intelligence (AI) and machine learning (ML) as innovative applications for predicting medical emergencies. In this paper, the use of Interpretable Machine Learning Regressors for Mild Hypothermia Prediction in General Surgical Operations was leveraged. Specifically, building, testing, and optimization of Extreme Learning Machine (ELM), Linear, Random Forest (RF), Logistic, and Support Vector Machine regression models were done where an accuracy of 98.76%, 98.79%, 98.69%, 73.28%, and 29.34% respectively was obtained upon model tuning and hyperparameter optimization. SHapely Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) based on physiological vitals were transparently provided. This work can contribute to Society 5.0 by improving patient outcomes of general surgical operations, reducing healthcare costs, and increasing the efficiency and effectiveness of Intelligent Healthcare Systems.