Health data preparation framework for unsupervised models (HealthDataPrep): A study with Alzheimer's disease mortality data from Brazil

Healthcare data preparation challenges often arise from its complexity and heterogeneity. This study introduces a framework for preparing structured and non-temporal healthcare data, tailored for unsupervised learning models. The HealthDataPrep framework extends CRISP-DM, guiding data preparation in four key areas: data selection, data cleaning, feature engineering, and data privacy. Validated using a dataset on Alzheimer's disease mortality in Brazil, the framework generated two datasets utilized in unsupervised algorithms, K-Modes for clustering and Apriori for association rules. Results revealed sociodemographic profiles vulnerable to the disease, predominantly featuring Alzheimer's mortality in women over 80, widowed, and with low education in the Southeast region. Furthermore, the study identified associated causes of Alzheimer's mortality, such as pneumonia, respiratory failure, and sepsis. These insights are crucial for shaping public policies and intervention programs aimed at reducing Alzheimer's mortality. The research contributes methodologically to healthcare data preparation and enhances our understanding of Alzheimer's mortality in Brazil.