Desenvolvimento de um modelo de predição de desligamento voluntário de colaboradores: uma abordagem multimodal / J

Abstract

This research developed a predictive model for forecasting voluntary employee turnover within one year in a digital platform operating in the e-commerce sector, within a oneyear period, using the CRISP-DM (Cross-Industry Standard Process for Data Mining) methodology. Through a comprehensive analysis of the data, demographic, organizational, and performance variables were used to train and test several machine learning algorithms. The modeling process included combining the results of three models to define risk levels, prioritizing the reduction of type II errors, which are critical in this context. The results highlight the effectiveness of the predictive models used. With F-beta Scores of 95.3%, 84.7%, and 82.2% for Random Forest, LightGBM, and CatBoost, respectively, the models were able to accurately estimate turnover probabilities, enabling the creation of a risk strategy integrated with compensation and retention decisions. The approach proved valuable for anticipating turnover and assisting HR management in implementing more targeted and effective retention actions. The study demonstrates how the application of machine learning techniques can add strategic value to people management and organizational decision-making.