Aplicação de técnicas de business intelligence em sistemas de apoio à tomada de decisão de produtores rurais

Abstract

Through partnerships between higher education institutions in the city of Bagé and the Brazilian Agricultural Research Corporation, Decision Support Systems are being developed that aim to assist the rural producer in making decisions on beef cattle. The difficulty arises when it becomes necessary to perform analyzes and establish relations between the variables of these systems automatically. In the way they are being developed, systems may end up supplying different versions of a true breeding herd, leading to greater difficulties. The purpose of this paper is to integrate data from two Decision Support Systems, using Business Intelligence techniques. The objective is to enable farmers to carry out analyzes based on the integration of environmental, economic and productive indicators data, and to provide means for integrated and simplified visualization of this information. For the implementation of the solution, only free and open source technologies were used. As results, a star schema dimensional model was elaborated, in which one fact table was modeled with economic, productive and environmental indicators of the cattle ranching. These indicators can be analyzed in time, by location, area ranges of properties, among other dimensions. The ETL process was carried out, allowing the integration of the data of the systems into a single repository, with a unique version of the truth of the different faces of the cattle ranching. The Saiku Analytics data visualization tool, integrated in the Pentaho suite, allows the analysis of OLAP-style information through tables and graphs, as well as allowing the export of data in reports and in spreadsheets. The solution can help answer questions such as "the average productivity of farms by location", "total costs of properties by area ranges", "variation costs and zootechnical indicators per year", among others. Several issues were detected, also presented in the literature review, in the processes that involve integration of data from heterogeneous sources. Problems of data granularity, systems developed without the integration bias, difficulties for a clear understanding of all aspects of the problem domain and no available legacy systems documentation were highlighted.