Examining Patients Length of Stay Estimation with Explainable Artificial Intelligence Methods

Forecasting the duration of patients stays is immensely beneficial for the administration of hospital resources. Numerous medical facilities globally, particularly in less urbanized regions, encounter restricted resources to house inpatients. Nevertheless, traditional hospital administration frameworks fail to estimate the duration of patient stays in the initial stages, resulting in several adverse outcomes for hospital operations. Using AI for patient admissions can help hospital management to better plan resources and deliver healthcare services more effectively. In addition, based on the predictions of the AI model to be created to predict patient hospitalization times, using XAI techniques to analyze which factors affect patient hospitalization times more and how they affect them can provide great benefits for hospital management. In this study, we used Microsoft's open source dataset on hospitalizations from Kaggle. In the study, GBR, DT, Ridge, LGBM, SVR, RF, XGBR, MLPR algorithms were used and GBR and XGBR algorithms gave the highest results with %96. The results of GBR, which is the complex classification algorithm selected among the highest performance results, were analyzed in three XAI frameworks with LIME, SHAP and ELI5. This research significantly aids healthcare professionals in understanding the model's forecasts better and utilizing it in hospital administration planning to enhance patient treatment and outcomes.