Improving diagnosis of acute coronary syndromes in an emergency setting: A machine learning approach

Abstract: Acute coronary syndrome (ACS) is the biggest people killer in the western world today. Despite well trained physicians and reliable diagnostic tools, diagnosing ACS early in the emergency departments (ED) remains a challenge. In this thesis we used machine learning, via logistic regression models and artificial neural network ensembles, to investigate the possibility of predicting ACS at an early stage using electrocardiogram data. Thorough comparisons were made to several expert physicians, currently working in the ED, to verify the models. In the context of neural networks we developed methods for the case based explanation of their decisions.