Predicting Hospital Stay Length Using Explainable Machine Learning

Abstract

Predicting how long patients will stay in the hospital is essential for better managing healthcare resources and improving patient care. This study dives into how we can use explainable machine learning techniques to estimate hospital stay durations, drawing from a dataset on Kaggle that includes various patient and hospital-related features. The main aim here is to create accurate predictive models while also shedding light on the factors that affect how long patients stay. We dive into a variety of machine learning algorithms, including Logistic Regression, Multi-Layer Perceptron, Random Forest, Gradient Boosting, and XGBoost.To assess how well each model performs, we use standard metrics like accuracy, precision, recall, and F1-score. On top of that, we employ explain ability tools like SHapley Additive exPlanations (SHAP) to help interpret the predictions and pinpoint the most important factors influencing the span of halt.