Knowledge Base for MENTAL AI, in Data Science Context

• Autor(a) principal: Muetunda, Faustino Paulo
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.6/13230
• topic: Algoritmo de Classificação; Dados de Saúde; Doenças Mentais; Inteligência; Saúde Mental; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Globally, 1 in 7 people has some kind of mental or substance use disorder that affects their thinking, feelings, and behaviour in everyday life. Mental well-being is vital for physical health. No Health Without Mental Health! People with mental health disorders can carry on with normal life if they get the proper treatment and support. Mental disorders are complex to diagnose due to similar and common symptoms for numerous types of mental illnesses, with a minute difference among them. In the era of big, the challenge stays to make sense of the huge amount of health research and care data. Computational methods hold significant potential to enable superior patient stratification approaches to the established clinical practice, which in turn are a pre-requirement for the development of effective personalized medicine approaches. Personalized psychiatry also plays a vital role in predicting mental disorders and improving diagnosis and optimized treatment. The use of intelligent systems is expected to grow in the medical field, and it will continue to pose abundant opportunities for solutions that can help save patients’ lives. As it does for many industries, Artificial Intelligence (AI) systems can support mental health specialists in their jobs. Machine learning algorithms can be applied to find different patterns in the most diverse sets of data. This work aims to examine and compare different machine learning classification methodologies to predict different mental disorders and, from that, extract knowledge that can help mental health professionals in their tasks. Our algorithms were trained using a total dataset of 3353 patients from different hospital units. These data are divided into three subsets of data, mainly by the characteristics that the pathologies present. We evaluate the performance of the algorithms using different metrics. Among the metrics applied, we chose the F1 score to compare and analyze the algorithms, as it is the most suitable for the data we have since they found themselves imbalances. In the first evaluation, we trained our models, using all the patient’s symptoms and diagnoses. In the second evaluation, we trained our models, using only the symptoms that were somehow related to each other and that influenced the other pathologies.