Data mining: predictive model of the socioeconomic impact of COVID-19 on the income of cooperative members validated using Neutrosophic TOPSIS and the Hausdorff Distance

Abstract

The COVID-19 pandemic generated a socioeconomic disruption that affected household income, making it crucial to identify modulating factors in vulnerable contexts. This study aimed to develop a predictive model using data mining to determine the socioeconomic, demographic, and well-being factors that influence the income level of 1,456 members of three cooperatives in Manabí, Ecuador, before, during, and after the lockdown. The Interindustry Standard Process for Data Mining methodology was employed. The survey data were segmented into training and test sets, with oversampling applied to the minority classes to correct for imbalances. Several supervised classification algorithms were trained and evaluated in this study. The results showed exceptional predictive performance, with neural networks achieving accuracies close to 90% and areas under the ROC curve greater than 0.95 in the test set. A negative impact of the lockdown was evident, with an increase in the number of members in the lowest income category. The most influential factors were the educational level, employment status, and perception of well-being. These findings offer cooperatives a tool to identify vulnerable profiles and design targeted support strategies, confirming the effectiveness of data mining in modeling economic vulnerability. Furthermore, validation using the Neutrosophic TOPSIS method and Hausdorff Distance corroborated the robustness of the predictive models by incorporating uncertainty and indeterminacy into the assessment of socioeconomic factors, strengthening the reliability of the results in crisis contexts.