Machine learning models for classification of myocardial infarction using the PTB-XL dataset
(2025) EEML05 20251Department of Biomedical Engineering
- Abstract
- Electrocardiogram (ECG) is an important tool for diagnosing myocardial infarction (MI), but analyzing the ECG signals can sometimes be difficult for physicians. Machine learning could potentially help provide a fast and consistent interpretation of ECG signals. This study aims to evaluate the performance of different machine learning models for classifying MI using ECG data from the PTB-XL dataset. Classical machine learning, deep learning, and ensemble models were trained, validated, and tested with a split of 70%, 15% and 15%. The classical machine learning models used extracted morphological features indicating MI as input while the deep learning model used raw data. In terms of accuracy the Convolutional Neural Network (CNN) performed... (More)
- Electrocardiogram (ECG) is an important tool for diagnosing myocardial infarction (MI), but analyzing the ECG signals can sometimes be difficult for physicians. Machine learning could potentially help provide a fast and consistent interpretation of ECG signals. This study aims to evaluate the performance of different machine learning models for classifying MI using ECG data from the PTB-XL dataset. Classical machine learning, deep learning, and ensemble models were trained, validated, and tested with a split of 70%, 15% and 15%. The classical machine learning models used extracted morphological features indicating MI as input while the deep learning model used raw data. In terms of accuracy the Convolutional Neural Network (CNN) performed best with an accuracy of 93% but in terms of ”Receiver Operating Characteristic - Area Under the Curve” (ROC-AUC), the best performing model, with an ROC-AUC of 97.52% was a stacking ensemble created with the CNN and the Support Vector Machine. The extracted features in the classical machine learning models were evaluated using SHAP values which indicated that T amplitude in lead II and ST deviation in lead III were the most important features across the classical machine learning models. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9205011
- author
- Österlund, Ruben LU and Rekstad, Hugo LU
- supervisor
-
- Donglin Liu LU
- organization
- alternative title
- Maskinlärningsmodeller för klassifisering av hjärtinfarkt med hjälp av PTB-XL datasetet
- course
- EEML05 20251
- year
- 2025
- type
- M2 - Bachelor Degree
- subject
- language
- English
- id
- 9205011
- date added to LUP
- 2025-07-01 09:32:44
- date last changed
- 2025-07-01 09:32:44
@misc{9205011, abstract = {{Electrocardiogram (ECG) is an important tool for diagnosing myocardial infarction (MI), but analyzing the ECG signals can sometimes be difficult for physicians. Machine learning could potentially help provide a fast and consistent interpretation of ECG signals. This study aims to evaluate the performance of different machine learning models for classifying MI using ECG data from the PTB-XL dataset. Classical machine learning, deep learning, and ensemble models were trained, validated, and tested with a split of 70%, 15% and 15%. The classical machine learning models used extracted morphological features indicating MI as input while the deep learning model used raw data. In terms of accuracy the Convolutional Neural Network (CNN) performed best with an accuracy of 93% but in terms of ”Receiver Operating Characteristic - Area Under the Curve” (ROC-AUC), the best performing model, with an ROC-AUC of 97.52% was a stacking ensemble created with the CNN and the Support Vector Machine. The extracted features in the classical machine learning models were evaluated using SHAP values which indicated that T amplitude in lead II and ST deviation in lead III were the most important features across the classical machine learning models.}}, author = {{Österlund, Ruben and Rekstad, Hugo}}, language = {{eng}}, note = {{Student Paper}}, title = {{Machine learning models for classification of myocardial infarction using the PTB-XL dataset}}, year = {{2025}}, }