Business Intelligence: Use of Data Mining Techniques for the Prediction of Internment Times

Abstract

With data collection increasing in clinical information systems, it became necessary to explore various technologies and methodologies to analyze this valuable knowledge. The objective of this investigation was to obtain an optimized predictive model of patient’s internment time in the Armed Forces Hospital, through the discovery of behaviors and patterns existing in the internment process, based on data mining techniques. The internments performed at the Armed Forces Hospital between 2013 and 2017, were the population target of this research and, in view of the objectives and the problem, the methodology that proved to be the most adequate was the CRISP-DM methodology. In the data preparation phase, 19 input attributes were selected, while in the modeling phase a regressive approach was applied with five regression techniques: Decision Tree, Naive, Multiple Regression, Random Forest, and Support Vector Machines. Random Forest was the best model, with a coefficient of determination 0.735 and predicting correctly 78.5% of the cases. Through a sensitivity analysis, was verified that the four most significant attributes, related to the patient’s clinical situation, contribute more than 50% to the explanatory capacity of the new model: Hospital Episode Type (31.9%), Physical Service (8%), Medical Specialty (7.5%), and Discharge Destination (7.2%).