“Evaluación de modelos predictivos supervisados basados en registros administrativos para el análisis del desempeño laboral en una pyme industrial ecuatoriana.”

Abstract

This article analyzes the application of artificial intelligence and data analytics in human talent management within an Ecuadorian industrial SME. To this end, administrative records of 55 workers corresponding to the second half of 2025 were utilized, considering three operational indicators: absences, tardiness, and production errors. These indicators were compared with semiannual job performance, evaluated on a scale of 1 to 100 by the Integrated Management System Coordinator. The study uses a quantitative approach based on descriptive, correlational analyses, and supervised predictive modeling through linear and logistic regression implemented in Python. The results show variability in the operational indicators and a concentrated distribution of job performance. The linear correlations between the indicators and performance were very weak (r between –0.072 and 0.022), suggesting the absence of a significant linear association. The linear regression model presented low explanatory power, while the logistic regression achieved an accuracy of 0.471—a value that does not exceed the level expected by chance in a classification with a relatively balanced class distribution, indicating limited discriminative capacity. Consequently, the specific hypotheses related to the predictive capacity of the operational indicators were not empirically confirmed in the analyzed sample. The findings reinforce the conception of job performance as a multifactorial phenomenon and suggest that isolated administrative metrics are insufficient for its explanation or robust prediction.