Aprendizagem de máquina e análise de componentes principais na avaliação de qualidade do solo com pastagens

Abstract

Soil can be analyzed from several perspectives, such as soil enzymes, β-glucosidase, cellulase, xylanase and amylase, which are important for the decomposition of organic matter, soil respiration, which involve microorganisms that release CO2 during decomposition, being a crucial process for the carbon cycle. To address this area in a mathematical model, one can consider measuring the rates of enzyme activity and soil respiration and correlating them with environmental and management variables, another element is the microbial quotient (qMic) and metabolic quotient (qCO2). Which evaluates the efficiency of microorganisms in using available organic carbon (qMic). qCO2, on the other hand, relates soil respiration to microbial biomass, being an indicator of efficiency in the use of organic carbon. This project aims to create a mathematical model that can involve the correlation of these quotients with environmental variables to understand the factors that influence microbial activity, chemical and physical attributes of the soil. In chapter 1, the main component analysis (PCA) model is worked on, which helps to reduce the complexity of the model and generate appropriate metrics such as F1-score, facilitating interpretation and processing. In chapter 2, to achieve maximum generalization and accuracy of results, unsupervised machine learning, Self-Organizing Maps (SOM), is used with the strategy of reducing the dimensionality of the matrix of values, the Random Forest (RF), to minimize the attributes, and in the last evaluation chain the ranking of these attributes, to better make available and form new views for interpretation. All data undergoes pre-processing submitted in each model seeking maximum generalization, so that it can cover the largest number of soil types, and maximum accuracy. In all models used, each characteristic is important in each model to identify which attributes should be maintained or discarded. By combining these approaches, it helps to optimize available resources and simplify the analysis and decision-making process.