Imbalanced data learning for biomedical application
Imbalance data learning is an area of study motivated by application of machine-learning concept on real-world data. Due to the overwhelming instances of majority class, conventional machine learning algorithms have poor performance on prediction for minority class instances. This report explores a...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
2011
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/44917 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Imbalance data learning is an area of study motivated by application of machine-learning concept on real-world data. Due to the overwhelming instances of majority class, conventional machine learning algorithms have poor performance on prediction for minority class instances. This report explores a method of imbalance data learning, investigating the use of HRV parameters and vital signs as predictor of cardiac arrest occurring 72 hours within hospital admission.
This project aims to present a series of steps, starting from the processing of ECG data to the classification of patients using various input features. Input features consist of 16 HRV parameters and 8 vital signs. Several machine learning algorithms were developed and integrated into a main package for the automatic classification of 857 patients. Results show that these algorithms are able to achieve a sensitivity of 64.44% and specificity of 63.79%. This means that there is 64.44% chance of labelling a positive patient as positive, and 63.79% chance of detecting a negative patient as negative. |
|---|