Modelo de clasificación de depresión o ansiedad en base a información no estructurada

Abstract

Abstract: The objective of this Degree Project is to detect depression and anxiety using Artificial Intelligence and natural language processing techniques, where models such as BERT(Transformers), RNN and CNN were evaluated, using a data set of 6982 Facebook posts that are labeled to identify these disorders. The BERT model was the best, reaching an accuracy of 99% in text classification, surpassing the RNN and CNN models because they presented difficulties in capturing contexts and semantic relationships in the messages. This choice was made by analyzing the evaluation metrics obtained by the models in the four experiments carried out, data balancing was carried out using SMOTE, and dataset cleaning such as text cleaning, elimination of stopwords and stemming was applied. The results obtained managed to accurately detect the problems of mental illnesses such as anxiety and depression, helping to provide effective and quick solutions in the area of mental health for more precise and early interventions.