Integration of Machine Learning and Multicriteria Modeling for Evaluation of Predictive Models for Diabetes

In the context of diagnostic medicine, the study in question establishes an integrated Data Science framework to explore patterns and predict the occurrence of diabetes from clinical observations of patients. The initial stage consists of the construction and organization of a database aimed at analyzing the relationships between variables associated with the diagnosis of the disease. Then, an exploratory analysis is performed with descriptive statistics and unsupervised machine learning techniques. In the predictive stage, different supervised models are evaluated, compared using stratified performance metrics. To support the choice of the most appropriate model, the Gaussian AHP multicriteria method is applied, which assigns weights to the criteria based on the variability of the observed metrics. The results allow us to identify the most favorable models for forecasting, in addition to discussing the feasibility and limitations of the proposed framework. At the end, the conclusions of the research and suggestions for future work are presented.