Técnicas de machine learning aplicadas no processo de certificação de granjas avícolas

Abstract

In recent years, the field of Artificial Intelligence (AI) has shown many advances, especially with the use of machine learning and neural networks, generating great research opportunities. A large part of the expansion of the AI area is due to the increase in the use of machine learning techniques, which aim to train computers to be able to make predictions and make decisions based on a set of data. This study aims to find patterns and apply machine learning techniques using a database from the Animal Health Defense Platform of Rio Grande do Sul (PDSA-RS) with the purpose of automating operations and/or assisting users to make more accurate and intelligent decisions in the bird certification process. The type of learning used is the supervised with a focus on classification, where the algorithm is trained through a set of data and tries to predict which class a new instance belongs to. The algorithms to be studied and used are (a) Support Vector Machine (SVM) which is excellent for binary and linear classification, but also with the possibility of multi-class classification with the use of the kernel trick. (b) K-nearest neighbors (KNN) which has a simple operation and a great performance for multi-class classification. (c) Decision Tree, being simple to understand how the generated model works, because, basically, it creates a decision tree based on rules that are learned by the algorithm in the training stage, facilitating human understanding of the model's operation and (d) Neural Networks which are a more robust type of algorithm, with a very large learning capacity. The application of these techniques was made using the Python programming language in addition with the libraries Scikit-learn and TensorFlow on a set of data in CSV format, extracted from a PDSA-RS database. The accuracy of the traditional algorithms reached a maximum of 79.88% with KNN, while with neural networks the result was 94%.