Data Mining for Hospital Morbidity Forecasting

摘要

Growing demand for hospital healthcare services has brought significant challenges for their managers. Variables with high uncertainty degree, such as the number of patients and the duration of their treatments, hinders the planning processes and make it difficult to properly comply with the established strategies. Controlling and identifying factors that affect the hospital management process depends on health database analysis. Therefore, it is important to consider the possibility of prospecting useful knowledge from the stored data. The objective of this research is to evaluate the hospital morbidity prediction through different data mining methods on ambulatory and hospital procedure records obtained from Brazilian public health databases. The research method consists of performing a predictive data mining by applying supervised learning algorithms on a regression problem. The highest Pearson correlation coefficient individually obtained in the three-month prediction time interval, through the data mining method that applied random forest associated with an attribute selection algorithm on the disease group of the ICD10 chapter XVI (Certain Conditions originating in the Perinatal Period), was 0.9682. Different results were achieved depending on the method applied, the group of diseases analyzed, and the proposed prediction time interval, which led to the conclusion that data mining on ambulatory and hospital records allowed the prediction of hospital morbidity. The hospital morbidity predictions obtained can minimize the undesired effect of the demand randomness for health services in the decision-making process.