Desarrollo de un modelo de inteligencia artificial para ayudar en la toma de decisiones en política pública aplicado a la vocación agrícola desde la perspectiva sociodemográfica en el territorio colombiano

Abstract

The main purpose of this doctoral thesis is to develop an artificial intelligence model as a mechanism for public policymakers to improve decision-making processes and enhance the agricultural vocation in Colombia, considering sociodemographic variables. Therefore, three main objectives were defined: first, the development of a conceptual model of the agricultural vocation; second, the design and implementation of the model; and third, the evaluation of the model. The first step included a conceptual revision of public policies and artificial intelligence, and a state-of-the-art revision. On the one hand, public policies related to the agricultural sector and employment in Colombia between 2006 and 2022 were analyzed; on the other hand, a systematic revision of artificial intelligence applied to research on agricultural policies was also carried out. An evolution was observed in the agricultural policies: the focus was rooted in the productivity of the sector, to later prioritize sustainable development. In addition to agricultural processes, the use of artificial intelligence has reached evaluation and prevision processes for efficient use of natural resources like water and land, aiming to anticipate economic and environmental impacts and promotion of sustainable development. After the conceptual review, which included a systematic revision of relevant scientific publications followed by a qualitative-quantitative analysis, a conceptual model and a state-of-the-art for the agricultural vocation was developed. The development of the mode revealed that the agricultural vocation embodies three essential perspectives for sustainable development: sociodemographic, economic, and environmental. Consequently, the model integrates such perspectives as well as technological and public policy variables to support public agrarian policy processes. During the design, the most significant findings show agrarian public policies must be comprehensive and include healthcare issues, education, security, and infrastructure in rural zones to guarantee efficient and sustainable production. The next step was developing the computational model, which included creating the database. The first step involved generating various indexes using the sociodemographic variables taken from the conceptual model. The sources of information were mainly governmental and an imputation process was carried out to complete the missing data. A correspondence analysis was completed using the database and the conceptual model as references to identify relationships between variables starting from the Colombian data. Thus, a more suitable computational design for the conditions set was established. From this design, more suitable techniques and configurations were implemented for both predictions and phenomenon analysis. The techniques employed included artificial neural networks, support vector machines, and neuro-fuzzy inference systems. One of the main findings taken from the model implementation revealed that public policies intended to promote agricultural employment must be based on promoting secondary school or higher education for the rural population; besides, agricultural employment must be formalized with open-ended contracts and decent wages. During the evaluation of the computational model, it was concluded that artificial neural networks are one of the most efficient techniques for making predictions, followed by neuro-fuzzy inference. Nevertheless, the neuro-fuzzy inference system proved to be the most suitable technique useful for understanding the phenomenon, and thus providing public policy guidelines focused on agricultural vocation.