Feature Selection for Heart Beat Classification
Author Information
Author(s): G. Doquire, G. de Lannoy, François D., M. Verleysen
Primary Institution: Catholic University of Leuven
Hypothesis
Can feature selection techniques improve the classification of heart beats in interpatient scenarios?
Conclusion
Using feature selection techniques, a small number of relevant features can significantly improve heart beat classification performance.
Supporting Evidence
- The best performance on the test set was achieved with only two features.
- The mutual information criterion effectively ranked features for selection.
- The study found that many commonly used feature sets did not contribute to classification performance.
Takeaway
This study shows that by picking the right features, we can make better guesses about heart beats, which helps doctors understand heart problems.
Methodology
The study used two feature selection methods: a wrapper approach with a weighted LDA classifier and a filter approach using mutual information with a weighted SVM classifier.
Potential Biases
Potential bias in feature selection could affect the generalizability of the results.
Limitations
The study may not generalize well to other datasets or classification tasks due to the specific features and methods used.
Participant Demographics
The study involved 44 patients from the MIT-BIH arrhythmia database.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website