Sistema de classificação do mel de Tetragonisca angustula por região através de parâmetros físico-químicos combinados com redes neurais artificiais

Abstract

Meliponiculture involves breeding social bees that inhabit colonies and possess atrophied stingers. This activity has a low initial cost and is an excellent alternative source of income, also arousing interest for its potential to pollinate native plants of agricultural interest. The species under study in the project, Tetragonisca angustula, is one of the cleanest bees and one of the most adaptable species in terms of nesting habits, which is characteristic of its good hardiness. Therefore, the present work aims to use artificial neural networks to classify stingless bee honey by physicochemical parameters. The selected parameters were moisture, pH, acidity, HMF, and ash content, with the choice based on a larger amount of data obtained compared to others. These were collected and organized by regions of Brazil (South, Southeast, Northeast, Midwest, and North), and later these values were transformed into a code model for reading by the Weka software. The classification algorithms chosen to be worked on were "Multilayer Perception," "Random Forest," "BayesNet," "KStar," and "LogitBoost," adopting a confidence level of 80% for the proposed neural networks classification, with the ideal values being close to 90%, but this is justified by the lack of homogeneity in the data, especially due to the scarcity of physicochemical information on the jataí bee in the North and Northeast regions of Brazil. The program proved efficient in the proposed objective, with the reading of two hundred and twenty-two total collected data, with three out of five selected classification algorithms showing more than 80% of adopted reliability (Random Forest with 82.88%, KStar with 84.23%, and LogitBoost presenting a value of 80.63%), showing a satisfactory fit to the available data. Therefore, this study can provide information to predict and understand the different compositions of Tetragonisca angustula bee honey, contributing to the development of more comprehensive regulations and quality standards adapted to the natural variations found in different geographical regions of Brazil.