Machine learning and econometric applications for increasing profitability and efficiencya case study on sustainable production and trade in agro-based industries

Résumé

By 2050, the world population is expected to reach 9 billion people. Global population growth will lead to an increase in consumer demand for products derived from agriculture. As a result, agricultural production is expected to increase by approximately 70% worldwide. Over the last years, technological applications have made it possible to increase agricultural production and contribute to reduce pollution, the sustainable use of natural resources, cost forecast, risk mitigation and even potential bankruptcy anticipation. In this regard, one of the global, near-future challenges is designing methodologies that enable companies in the agricultural sector to be more efficient and sustainable regardless of the technological development index of the region in which they are located. In this research, taking the agricultural sector as a case study, has been designed and validated a modular methodology that combines machine learning algorithms and econometric models aimed at improving the management of resources, allowing companies to be more competitive and sustainable in order to encourage their investment. To develop the methodology, three experiments were carried out. The first experiment was aimed at measuring resource efficiency based on a non-parametric method for estimating production frontiers in which the costs derived from data transfers were incorporated, making it possible to identify the most optimal production frontiers taking into account technological costs. For the second experiment, a multi-agent system has been designed to predict price variations in futures markets for agricultural products. The multi-agent system is been designed as a decision support system in which potential buyers or sellers can incorporate environmental impact parameters. Finally, the last experiment consists of the design of a case-based reasoning methodology for the recommendation of investment in a company. The last experiment enables the incorporation of investment of capital to companies in the agricultural sector. In addition, to increase the performance of the investment recommender system an improvement has been implemented in the third experiment. This improvement has made it possible for the system’s classification element to compare different evaluation metrics in situations where the data labels are not balanced.