Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/95CF8E43-6FDB-4782-A41C-DFAEAC76BB26 | ||
| Year | 2025 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Stavros Flourentzou, "The use of machine learning algorithms in predicting patient outcomes with heart failure", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2025 https://doi.org/10.26233/heallink.tuc.104045 | ||
| Appears in Collections |
The weakening of the heart muscle is a major cause of illness and death all over the world, so there is an immediate need to have precise predictive tools, which can enable doctors to make well-informed decisions and help them to use healthcare resources appropriately. Recent advancements in artificial intelligence and machine learning have improved the capability to predict outcomes in heart failure patients by utilizing numerous and high-resolution healthcare datasets. The research of this diploma thesis considers the application of different types of machine learning algorithms (Logistic Regression, Support Vector Machine, Random Forest, Extreme Gradient Boosting, Light Gradient Boosting Machine, Tabular Transformer) to precisely predict critical outcomes, including intensive care unit (ICU) admission and mortality within the first month of hospitalization. The useful data that these models are using were extracted from a well-known medical database, Medical Information Mart for Intensive Care IV (MIMIC-IV). Experimental results demonstrate that the proposed models effectively predict ICU admissions with high accuracy and robust performance metrics. However, the prediction of mortality within one month after hospitalization demonstrates limited effectiveness due to significant class imbalance, leading to suboptimal performance in area under the curve (AUC) and accuracy. Despite applying class balancing techniques, the model struggles to accurately identify minority class instances. These findings underscore the challenges of class imbalance in prediction problems and the need for more advanced resampling or algorithmic approaches to improve predictive accuracy. The proposed models are designed to improve prognostic accuracy and identify high-risk patients, ultimately contributing to personalized treatment strategies and better healthcare management.
Copyright © DIAS 2013
