Dissertação de Mattyws Ferreira Grawe

Aluno: Mattyws Ferreira Grawe
Orientadora: Profa. Dra. Viviane Pereira Moreira

Título: Heterogeneous Ensemble Models for In-Hospital Mortality Prediction
Linha de Pesquisa: Mineração, Integração e Análise de Dados

Data:  30/09/2021
Horário: 13h30min.
Esta banca ocorrerá excepcionalmente de forma totalmente remota. Interessados em assistir a defesa poderão acessar a sala virtual através do link: https://mconf.ufrgs.br/webconf/00149248

Banca Examinadora:
– Prof. Dr. Cristiano André da Costa (Unisinos)
– Profa. Dra. Mariana Recamonde Mendoza Guerreiro (UFRGS)
– Prof. Dr. Joel Luis Carbonera (UFRGS)
Presidente da Banca: Profa. Dra. Viviane Pereira Moreira
Abstract: The use of Electronic Health Records data have extensively grown as they become more accessible. In machine learning, they are used as input for a large array of problems, as the records are rich and contain different types of variables, including structured data (e.g., demographics), free text (e.g., medical notes), and time series data. In this work, we explore the use of these different types of data for the task of in-hospital mortality prediction, which seeks to predict the outcome of death for patients admitted at the hospital. We built machine learning models for each data type and combine them into a heterogeneous ensemble model using the stacking strategy. By applying deep learning algorithms of the state-of-the-art in classification tasks and using their predictions as a new representation for our data we could assess whether the classifier ensemble can leverage information extracted from models trained with different data types. Our experiments on a set of 20K ICU stays from the MIMIC-III dataset have shown that the ensemble method brings an increase of three percentage points, achieving an AUROC of 0.853 (95% CI [0.846,0.861]).
Keywords: Mortality prediction. data types. machine learning. ensemble. time-series.