- Accuracy is 85% without implementing any feature selection techniques as opposed to 88% accuracy with feature selection [5].
- Model rarely predicts 10-year CHD risk above 50% which is indicative of the training set used which has majority data with negative CHD label.
- Model can differentiate low-risk from high-risk patients.
- Some significant variable requiring interventions are smoking, cholesterol, systolic blood pressure and glucose.
- Overall model could be improved by using data with more variance.
Firstly, the current results of the model are limited by the small amount of data and the missing data in some features. Incorporating more accurate and precise patient instances may improve the accuracy results. Secondly, we can try to improve the classifier to be able to work on real-time data such as ECG signal and predict CHD risk. Lastly, other models can be implemented to find out which one has the best overall performance and higher accuracy.