Abstract
Hospital length of stay (LOS) of patients is an important factor for planning and managing the resource utilization of a hospital. There has been considerable interest in controlling hospital cost and increasing service efficiency, particularly in stroke and cardiac units where the resources are severely limited. This study introduces an approach for early prediction of LOS of stroke patients arriving at the Stroke Unit of King Fahad Bin Abdul-Aziz Hospital, Saudi Arabia. The approach involves a feature selection step based on information gain followed by a prediction model development step using different machine learning algorithms. Prediction results were compared in order to identify the best performing algorithm. Many experiments were performed with different settings. This paper reports the performance results of the two most accurate models. The Bayesian network model with accuracy of 81.28% outperformed C4.5 decision tree model (accuracy 77.1%).