An unsupervised machine learning approach for heart disease prediction
Heart disease, also known as cardiovascular disease, persists as a primary cause of mortality on a global scale, necessitating effective prediction methods. This study introduces a novel modelling approach utilising the Self-Organising Map (SOM), an unsupervised machine learning approach, for heart...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Year Project / Dissertation / Thesis |
| Published: |
2024
|
| Subjects: | |
| Online Access: | http://eprints.utar.edu.my/6840/1/AM_2106963_Final_Lim_Yu_Jiun.pdf http://eprints.utar.edu.my/6840/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Heart disease, also known as cardiovascular disease, persists as a primary cause of mortality on a global scale, necessitating effective prediction methods. This study introduces a novel modelling approach utilising the Self-Organising Map (SOM), an unsupervised machine learning approach, for heart disease prediction with the incorporation of Particle Swarm Optimisation (PSO), a
metaheuristic optimisation algorithm. The SOM model was employed to analyse and cluster patient data based on intrinsic patterns without requiring predefined labels, allowing for the identification of individuals with heart
disease. The results demonstrated the SOM’s capability in distinguishing between healthy and diseased individuals, offering a robust approach for early detection of heart disease. By integrating PSO to fine-tune SOM’s
hyperparameters, the SOM model achieved superior predictive performance, with an accuracy of 94.44%, precision of 100%, recall of 92.86%, F1-score of 96.30%, and a quantisation error of 0.024. Furthermore, this study explored the
impact of SOM's visualisation techniques, such as heatmaps and the Unified Distance Matrix (U-Matrix), on the comprehension of cardiovascular conditions. The U-Matrix of the optimised SOM model provided insightful visualisation that revealed two distinct clusters, effectively illustrating the health status of individuals with similar health conditions concerning heart disease. These visualisations afford a more profound understanding of the
hidden relationships within the heart disease data, enhancing the model's interpretability and facilitating a better management of heart disease. The findings suggest the potential of integrating SOM into clinical workflows,
offering a potent tool for healthcare professionals in the fight against heart disease.
|
|---|