Machine Learning Based Early Sepsis Prediction for Diabetic Foot Patient

The early and accurate prediction of sepsis, a potentially fatal illness marked by a dys-regulated host response to infection, is crucial for improving patient outcomes. The study focuses on developing a predictive model that can identify the likelihood of sepsis in Bangladeshi diabetic foot patients, rather than providing a definitive diagnosis. This study utilizes a real-world cross-sectional dataset and explores the performance of various machine learning models in predicting sepsis using a comprehensive set of features derived from domain knowledge and proposing a missing value imputation method utilizing automated machine learning concept through Pycaret. The proposed method, IKMHybridImputer is a combination of k-means clustering and iterative imputation. The study utilizes z-score normalization and to address the challenge of missing data, performance of two imputation methods: mean and IKMHy-bridlmputer approach have been compared. Staratified K-fold cross validation is applied to all the thirteen machine learning models implemented. Finding the best model among all of them, optimization for hyper-parameter tuning have been tested out. The results showed that the non-tuned Adaboost model achieved the best overall performance using the proposed method. To gain deeper insights into the performance of the best model, the study employed SHAP to explain the predictions.