Establishment of a Mortality Prediction Model in Critically Ill Influenza Patients by Machine Learning
| Autor: | Yen-Chun Fang, 方彥均 |
|---|---|
| Rok vydání: | 2018 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 106 Influenza is an acute respiratory disease that causes 250,000-500,000 deaths worldwide each year. Outbreak of the influenza is characterized by the rapid spreading in the epidemic areas, and such a rapid spreading may lead to an abrupt increase of subjects with severe influenza requiring the intensive care within weeks during the outbreak. Taiwan experienced a catastrophic influenza outbreak in the spring of 2016, and approximately 120 subjects died from severe influenza during the outbreak. A practical mortality prediction model is needed for allocation of resource during the influenza outbreak. However, current mortality prediction parameters in critically ill patients are designed for general critical illnesses, and the mortality prediction model for critically ill influenza patients is still lacking. Therefore, we used the data set obtained in the eight medical centers in Taiwan during the influenza outbreak and aimed to establish a mortality prediction model for critically ill influenza patients through application of different machine learning algorithms In recent years, a number of machine learning algorithms have been used by physicians to diagnose cancer or diabetes-associated retinopathy and to predict mortality through automatic interpretation of medical images or electronic medical records, text mining technique, and various machine learning methods. The purpose of this thesis is to establish a predictive model that can be used to develop a practical model to predict the 30-day mortality of critically ill influenza patients through machine learning technology. This data set has 336 patients, including 264 survivors and 72 deaths. We selected key features through discussion with professional doctors and Sequential Backward Selection (SBS) algorithm, handling Unbalanced Data Problems with Random Over Sampling and Synthetic Minority Over-sampling Technique, the predictive models are built using Decision Tree (DT), Random Forest (RF), Linear Support Vector Machine (Linear-SVM) and Non-linear Support Vector Machine (Non-linear SVM), Compare the results of model predictions. After the oversampling process, the prediction results are more obvious. This thesis esthe tablishes a prediction model for mortality in critically ill influenza patients and compares the efficacy between models. In the future, a decision support system can be established to assist physicians in the initial diagnosis. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|