Riskwatch: A Model for Improved Preoperative Risk Assessment of Anesthesia in Medical Science Using Machine Learning

Anesthesia is a medical breakthrough with certain inherent risk factors. This paper presents a system for preoperative assessment of risks associated with anesthesia. To ensure patient safety, managing these risks is a top priority for anesthesiologists. An ever-increasing ratio of the number of surgeries to the number of doctors can inevitably lead to errors in risk judgment. This paper classifies patient data into four risk categories: due risk, low risk, moderate risk, and high risk based on supervised machine learning (ML) models. The data preprocessing technique of binning is applied to segregate the data. Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) is used to compare various supervised ML methods such as K nearest neighbors (KNN), logistic regression (LR), support vector machine (SVM), gradient boosting (GB), decision tree (DT), random forest (RF), and an ensemble model for multi-class classification. Explainable AI techniques like SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) offered valuable insights into the model’s decision-making process. The study compares ML models trained separately on raw and binned datasets, showing significant improvements in predictive accuracy with binning. The ensemble model achieved 95% accuracy in classifying the test patients using binned data. Our research introduces a ML-based system that can reduce preoperative risk and help anesthesiologists by providing real-time information, thereby enhancing patient safety, and reducing the likelihood of errors in preoperative risk judgment.