Desenvolvimento de um Modelo para Apoio à Tomada de Decisão na Gestão dos Gastos da Marinha do Brasil

Abstract

All around the world, governments face an endless challenge to promote an increasing number of public services with the same (or even less) volume of spending each fiscal year. Improving the efficiency of public sector spending means enhancing the state's capacity to act, as well as ensuring that it responds adequately to those who finance it: its citizens. In this context, information and communication technologies are the foundations for many of the innovative ideas that allow to reduce costs and, at the same time, promote the quality of the service provided. However, the innovation literature in the public sector points out that little attention was paid to innovations related to technological processes. As it is a field of study with the potential to facilitate data-driven decision-making, this research proposed to carry out an approach based on machine learning for the development of a model capable of supporting the decision-making process by understanding the variation of common and essential expenses in the more than 400 Brazilian Navy installations, responsible for the execution of several and different activities necessary to fulfill its institutional mission throughout the Brazilian territory. In order to achieve this objective, secondary data, coming from different sources and representing different attributes of the facilities, were adequately collected, treated, and loaded into an architecture that enabled the learning process from different perspectives. Subsequently, the best models went through a procedure capable of identifying the most relevant attributes based on the strength with which they contribute to the predictive performance of the model. The results show that, despite the data limitations, the model has achieved a reasonable capacity for forecasting, and the knowledge of the most important attributes that contribute to it are capable of generating useful insights that assist in the managerial decision-making process related to the allocation of resources. The performance achieved also maintains that the developed architecture can also be considered an additional result of this study. Its implementation as a process on the organization's data analysis platform will enable a more in-depth analysis of the problem presented, with a larger volume of data and attributes, as well as its reuse in other areas of interest.