Classificação de risco de extinção de vertebrados utilizando relatórios de texto

Abstract

In a society characterized by continuous growth, increasing population, technological advancements, and rapid adaptation, the changes accompanying this growth can have profound impacts on various areas of the planet. This project focuses on exploring the impacts of human carnivorism on vertebrates and analyzing how the changes made by humans to meet their dietary needs, from both animal and plant sources, have affected vertebrates. The project’s objective was to develop a model capable of learning to classify species based on the type of extinction risk they face, using a manually categorized dataset. The primary focus of the results was to maximize the accurate identification of truly threatened species, thereby minimizing the number of false negatives. Throughout the project, three main classification models were constructed. The first model utilized a dataset with multiple labels, consisting of a total of 9 labels. Features were extracted from the dataset’s 25 columns of binary values representing the present threats. In the second model, while retaining the same labels, the features were replaced. Text data processing techniques were employed to generate new features, which were then used for training the second model. In the third and final model, the features from the first two models were combined for training, resulting in improved performance compared to the earlier models. This improvement was evident in both the aggregated values of all labels and the majority of individual labels, particularly those that initially exhibited unsatisfactory results. The final model yielded satisfactory results, with an average f1 score of 0.82 and an accuracy of 0.77 in identifying threatened species, indicating significant progress towards achieving the project’s objectives. Overall, the obtained results were considered satisfactory.