Explaing portuguese's public administration absenteeism through data mining

• Autor(a) principal: Costa, Leandro Miguel Bartolomeu da Cruz
• Data de Publicação: 2018
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/18298
• topic: Absenteeism; Portugal public administration; Data mining; RFM; Informática aplicada à gestão; Gestão de recursos humanos; Administração pública; Absentismo; Algoritmo; Análise de dados

Portuguese Public Administration (PPA) is the largest contractor in the country, with 12.8% of the Portugal’s active people working for it. Absenteeism and productivity are mutually connected. Thus, companies from public and private sector should always have it in mind, to prevent flaws in the processes and profit loss. Effectively, the main goal of this study is to understand PPA’s absenteeism, particularly the duration of the worker’s next absence, what leads to it, as well as explaining it, by creating a data mining model that fits the problem. To study PPA’s absenteeism it was collected data from a Human Capital Management (HCM) system, by extracting the annual absenteeism report, for 2016, and queries to the worker’s profile, absenteeism history and job characteristics, resulting in around 59,000 different absence records. Data mining techniques were used to clean the dataset and Recency, Frequency and Monetary (RFM) value methodology to add new variables to the problematic, originating richer information about the worker and the absence itself. Thereafter, the Support Vector Machines (SVM) algorithm was applied for modeling the absence duration in day and a 10-fold cross-validation scheme was adopted to assess and confirm the model’s robustness. Finally, major findings were revealed by this study as features related to the worker’s profile are less relevant than absence related features; the influence of the RFM methodology in this study, which managed to get all its computed variables in the 25th most important features; and the discovery of the most concerning employee profile.