Utilização do banco de dados internacional de testes de acidente vascular cerebral para aplicação da metodologia knowledge discovery in databases

Abstract

According to the Strategic Action Plan for Confronting Chronic Non-Communicable Diseases (NCDs) in Brazil 20112022 (2011), as previous CNCDs, the health problem of greater derivation and due to more than 70% of the causes of deaths in Brazil. Brazil. Observing the breadth of health in its macro form, the present work made a cut about the cerebrovascular accident (CVA), with the objective of classifying the patients that can evolve to death in up to six months through data mining techniques. Based on this assumption, and taking as a reference the International Bank of CVA Tests, which includes 3034 computed and 266 attributes, correlated works were identified, which in turn acted as supporting support in the choice and definition of attributes with the purpose to predict patients diagnosed with stroke that may progress to death. Four works guided the choice of attributes, applying them as decision tree and random forest techniques in such a way that they were applied as techniques for four distinct data sets. In this way, it was possible to measure the classification performance, as well as evaluate the attributes that best fit the proposed prediction. The best result was the one that had as reference the attributes used in the study by Dheepitha Babu et., Al (2021), with 69.67 of accuracy, precision of 0.749, followed by sensitivity of 0.794. The selected studies had the role of supporting the attribute selection process, on the other hand, a sensitivity in the health area demands better results. Thus, it is recommended to use the base used in this work, as well as the mentioned metrics, with the exception of having references to other attribute and techniques, aiming at better results.