Application of data mining techniques in a predictive model aimed at analyzing the effect of the COVID-19 pandemic on the income of cooperative members

Abstract

The health crisis caused by COVID-19 caused a significant socioeconomic disruption that reduced the income of numerous households, making it essential to identify the modulating factors in vulnerable populations. The purpose of this study was to design a predictive model, based on data mining, to identify the socioeconomic, demographic, and well-being determinants that influence the income level of 1,456 members belonging to three cooperatives in Manabí, Ecuador, during the periods before, during, and after the lockdown. The Cross-Industry Standard Process for Data Mining (CRISP-DM) was used for the analysis. The data collected through surveys were divided into training and validation sets, and oversampling techniques were applied to the underrepresented classes to balance the distribution. Subsequently, various supervised classification algorithms were trained, evaluating their comparative performance. The results showed highly satisfactory performance: the neural networks achieved accuracies close to 90% and ROC curve areas above 0.95 in the test set. It was found that the lockdown had a negative impact, reflected in an increase in members in the lowest-income category. Among the factors with the greatest weight in the prediction were academic education, employment status, and perception of well-being. The findings constitute a practical tool for cooperatives to identify more vulnerable profiles and develop targeted support strategies. Furthermore, the validation of the models using the Neutrosophic TOPSIS method and the Hausdorff Distance metric reinforced their robustness by integrating indeterminacy and uncertainty into the analysis of socioeconomic factors. This increases the reliability of the results in crisis scenarios, confirming the relevance of data mining for the study of economic vulnerability.