Inteligência computacional na previsão do absenteísmo e identificação de tendências absenteístas

Abstract

Absenteeism is considered a phenomenon defined as the non-attendance of the employee to work in a habitual way, with regular frequency and therefore the noncompliance of the obligations, as scheduled. Understanding and treating the causes of absenteeism has been a challenge, given the dimension of the phenomenon that encompasses psychological, physical and environmental causes. The prediction of absenteeism and the identification of absenteeism tendencies are important to reduce losses for the company and at the same time improve the quality of life of the employee. To this end, it is necessary to extract knowledge from databases that store information about employees of the company for several years, which opens space for the application of computational intelligence techniques, such as artificial neural networks. Thus, the objective of this work was to apply computational intelligence techniques in the prediction of absenteeism and in the identification of absenteeism tendencies. The database used is composed of 50 attributes with 2,403 medical license records from 39 employees collected during the period from January 2008 to December 2017. The computational experiments were carried out in two phases: Phase 1, called prediction absenteeism was In Phase 1, the artificial neural network of the type Multilayer Perceptron (MLP) was applied in Step 2 and in Step 2 the Rough Sets Theory was applied to reduce attributes using two reduction methods, the Genetic Algorithm and the Johnson Algorithm, and then applied the Multilayer Perceptron. In Phase 2, called the Self-Organizing Map artificial neural network, called Step 3. The comparison between the results obtained in Steps 1 and 2 made it possible to verify that the MLP presented the slightly better experimental error of the that the MLPs applied in the database reduced with the Rough Sets Theory. However, there was a considerable reduction in the processing time of the computational experiments in Step 2. It is noteworthy that the results of the two steps pointed positively to the prediction of absenteeism. In Phase 2, Step 3, identification of absenteeism tendencies with the Self-Organizing Map, the results generated also pointed positively to identify absenteeism tendencies by means of clustering evaluation. It is concluded that the computational intelligence techniques applied for the prediction of absenteeism and the identification of absenteeism tendencies have managed to reach the proposed objective and are presented as important techniques for the understanding and possible solution of this complex problem that afflicts both organizations employees.