Técnicas de mineração de dados para análise de uma base de pacientes com câncer de esôfago no período de 1998 a 2017

Abstract

Cancer is a disease responsible for thousands of new cases per year, so data are collected and stored to track cases and patients of the disease. In Brazil this data is stored in a national database called hospital cancer records. Health data has gained interest for use in knowledge discovery in databases, so this work aimed to apply data mining techniques to identify relevant patterns and profiles of patients diagnosed with esophageal cancer, since early diagnosis is essential. Mining rules of association, classification and grouping of data and applying methods of variable selection in variables related to factors prior to diagnosis, with the help of WEKA tool and the database available in the hospital cancer registry integrator. As a result, the apriori algorithm mined association rules, but was eventually influenced by the predominant values of some variables. The classification algorithms, J48 and reptree were tested with several configurations and achieved a satisfactory hit percentage, about 71%, by correctly classifying instances for the class referring to the primary tumor location. Data grouping algorithms, kmeans and especially EM, showed good performance when grouping by evaluating clusters against class when used mainly without the resample instance filter. The applied methods were able to classify and group the esophageal cancer cases, as well as revealing interesting patterns in the relation of the variables selected by variable selection methods.