Previsão do comportamento dos empregados no trabalho: mineração de dados aplicada em base de dados antropométricos, ergonômicos, absenteístas e presenteístas

Abstract

Absenteeism is a phenomenon defined as the employee's failure to show up to the workplace in a regular and regular manner; therefore, it is the non-fulfillment of obligations, as scheduled. Presenteeism, on the other hand, indicates the presence of the employee, albeit ill, in the workplace, however, the performance of his activities and functions may occur in an unproductive way. In this sense, anthropometric and ergonomic data, which are body measurements, are important when related to absenteeism and presenteeism, especially in activities classified as heavy work and with a high rate of repetitive activities. Predicting employee behavior is important to reduce losses for the company and improve the quality of life at work. Given this scenario, Data Mining techniques and some areas of Artificial Intelligence can be applied in predicting employee behavior at work. Thus, the objective of this study was to investigate the application of data mining, based on anthropometric, ergonomic, absenteeist and presenteeist data, to help predict the behavior of employees in the work environment. The computational experiments were developed in three stages in order to predict the behavior of employees - behavior that can be classified as presenteeist, normal and absenteeist. To carry out the experiments, two different architectures of artificial neural networks were applied: the Multilayer Perceptron (MLP) and the Radial Basis Function (RBF). In addition, Random Forest (RF) and Ant Colony Optimization (ACO) were also used. To enrich the experiment, three databases were used. Absenteeism and presenteeism data are common to the three databases and consist of 2,403 medical leave records from 39 employees collected during the period from January 2008 to December 2017. 10 attributes were considered in the first base, 11 in the second, and 25, on the third base. With the exception of the first base, the second and third were enriched with anthropometric and ergonomic data. The results showed better performance after the enrichment of the databases through MLP and the SOM network. In relation to the results of the computational experiments in Step 1, MLP obtained a hit rate of 99.91%, RBF, 97.08%, RF, 99.91%, and ACO, 80.65 %. In Step 2, the MLP obtained a hit rate of 99.91%, the RBF, 97.25%, the RF, 99.91%, and the ACO, 84.44%. In Step 3, the MLP obtained a hit rate of 99.96%, the RBF, 96.25%, the RF, 99.91%, and the ACO, 91.80%. Regarding processing time and performance, RF stood out as being the most recommended technique to assist in the prediction of presenteeist, normal and absenteeist behaviors in the work environment.