Software study to support agricultural production based on multicriteria analysis and machine learning

Abstract

Many regions of the country have long been extensively engaged in agriculture and livestock farming due to the relief, soil and climate conditions. The main environmental problem generated by these activities is soil degradation and this condition worries because it changes ecosystem structures and functions, generating serious risks to soil maintenance, imposing limits on its sustainability and reducing its productivity. In addition to soil physical and chemical conditions, global climate change has changed the temperature and rainfall regime, which directly impacts the quantity, quality, profitability and sustainability of agricultural production. The present project will develop a study for the construction of software that will indicate to agricultural producers if the soil of the property is degraded or not to the chemical and physical standards appropriate to the development of each culture through the generation of an Environmental Sustainability Index to ensure that ownership is sustainable over time. The objective is to combine analytical data with the knowledge and experience of the producer to provide new insights and improve decision making. In addition, this study provides data collection to feed a Machine Learning environment, which will ensure the efficiency of the whole process, making recommendations for improvements based on future projections from the combination of criteria. Such software can be used as a "property auditor", as well as indicating if the property's soil is degraded, it can "certify" the property as being sustainable when generating the Environmental Sustainability Index, improving the conditions for obtaining incentives and rural credit, for example. Existing solutions on the market are based on analytical data provided by precision agriculture, but so far no model has been found that encompasses objective (need) and subjective (experience and knowledge) values of the decision maker, providing recommendations according to context of decision. As a methodological proposal, the definition of the chemical and physical criteria of the soil that will be part of this study was obtained using the multicriteria analysis, specifically, the MACBETH socio-technical process, which provides details of the scenario under study, providing relevant information for decision making. it is a constructivist approach that encompasses the social and technical component. Machine Learning is a set of rules and procedures that allow computers to make suggestions and improve such suggestions over time when exposed to new data. Therefore, it is concluded that the major contribution of this project, taking into consideration the proposed methodology, will be to combine the subjective aspects with operational data to provide recommendations from future projections within a specific context and, with this, to minimize losses (sustainable use of soil) and improve financial autonomy (guarantee of agricultural production), as the structuring of the data from the multicriteria analysis combined with the "Machine Learning" feature will which can be potentialized or avoided. In the future, other criteria such as climate and disease could be included in the analyzes, improving local sustainability. (AU)