Heart Disease Prediction Using Tree Structured Parzen Estimator Optimization

Heart diseases involve diseases of the heart and blood vessels like coronary artery disease, arrhythmias and congenital heart diseases. Electrocardiogram or ECG is a test that assesses the activity, rhythm and electrical conduction of the heart and is paramount in diagnosing heart disorders. It should, however, be noted that even though the ECG is important in heart conditions diagnosis, it cannot by itself be a heart disease predictor, but still crucial in determining the health of the heart. TPE is a type of Bayesian optimization algorithm that is predominantly applied in hyperparameter tuning of predictive models. TPE can efficiently search the hyperparameter space as it creates a probabilistic model of the objective function and at every iteration pushes the most optimal hyperparameters and therefore increases the entire model's performance. This approach acts to augment the effectiveness of the models, particularly in complex prediction tasks. Convolutional Neural Networks (CNNs) are multi-layer perceptron neural networks used for images recognition, hence composed of filter layers, pooling layers, and fully connected layers. Their advantages include the ability to capture spatial features making them powerful in image classification and object detection tasks. The combination of TPE optimization with CNNs results in improved performance because hyperparameter tuning is automated and an optimal setting which maximizes the accuracy all that is required, with values as high as 87.3% accuracy in heart disease prediction.