Mineração de dados aplicada á predição de casos de abandono no tratamento de tuberculose em populações privadas de liberdade.

Abstract

According to the Brazilian Ministry of Health (MS, 2021), tuberculosis (TB) is an infectious disease that registers approximately 70 thousand new cases each year in Brazil, leading to the death of approximately 4.5 thousand people per year. A large number of these deaths are due to the high rate of abandonment of treatment, with an aggravating factor being the high rates of abandonment of treatment in cases involving populations deprived of their liberty (PPLs). In view of this problem and with the evolution of data mining algorithms, this work proposes the application of the stages of the KDD process (Knowledge Discovery in Databases) in the search for the best computational model for predicting cases of abandonment of TB treatment of PPLs. To perform these steps, the SEMMA methodology was used (Sample, Explore, Modify, Model, Assess) through cleaning techniques, pre-processing, attribute selection and application of data mining algorithms on a data set containing the records history of patients affected by TB registered in the Notifiable Diseases Information System (SINAN). The results obtained were evaluated according to their accuracy, precision, sensitivity and specificity. The main result was presented by the Random Forest algorithm, which obtained 97.07% accuracy in predicting abandonment in the treatment of TB in PPLs. To achieve the objective of this study, two manuscripts were written, the first being a systematic review of the bibliography related to the topic of data mining in the health area and the second the detailed description of the application of the SEMMA methodology adopted for the construction of the prediction models of TB treatment dropout in PPLs.