Uncertainty estimation of mental disease classification using Neural Networks
| Autor: | Rina García, Verónica |
|---|---|
| Přispěvatelé: | Høgskolen i Oslo, Halvorsen, Pål |
| Jazyk: | angličtina |
| Rok vydání: | 2022 |
| Předmět: |
Neural networks (Computer science)
Malalties mentals--Classificació Mental illness--Classification incertesa Uncertainty Mental Diseases Classification classificació de malalties mentals Xarxes neuronals (Informàtica) Informàtica::Intel·ligència artificial [Àrees temàtiques de la UPC] Bayesian Neural Networks Xarxes Neuronals Bayesianes |
| Popis: | The thesis goal is to investigate methods for medical time series data analysis and explore how uncertainty can be determined over a class prediction, specifically in the task of identifying in people the presence or absence of a particular disease from their data provided. We based this process on Bayesian Neural Networks (BNN) and on three open datasets containing the motor activity data of healthy participants and patients with depression, ADHD, or schizophrenia, named DEPRESJON, HYPERAKTIV, and PSYKOSE. We obtained the results with respect to several approaches of data treatment and several methods of feature selection. Moreover, a homogeneous style for the used open datasets has been established and it has been purposed as a way to measure uncertainty by calculating the entropy over the Softmax probabilities of the last layer of the model. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|