A Novel Computational Model to Predict Healthcare Problems Through Machine Optimization and Hyperparameter Tuning Logic

Heart disease (HD) remains one of the leading causes of mortality in the world and there are pressing needs to develop effective predictive models of HD at the early stages of development. The given research suggests a new type of computational framework combining optimization-based machine learning with innovative preprocessing and optimization approaches to the design of models predicting the presence of heart disease. It is an analysis of the data collection having clinically meaningful patient attributes. First, the dataset is cleaned with the help of the Discrete Wavelet Transform (DWT) to minimize noise and normalize signals to provide data of the necessary quality and make them more ready to use in a model. In feature extraction, Principal Component Analysis (PCA) algorithm is used to diminish dimensions and maintain most important variables in the realm of heart disease decision. After feature extraction, the data undergoes a Neural Network model which is optimized by Real-Parameter Numerical Optimization Algorithm (ROA). Since ROA has a high convergence rate and stability in locating global solutions, it is selected. As a way of enhancing the model, hyperparameter tuning- Bayesian optimization, is used to increase the overall predictive accuracy. Key performance indicators of the model include: accuracy 97%, precision 95%, Sensitivity 94% and specificity92%. The provided system shows good performance in all of them, which proves its potential in achieving robust and early prediction of heart diseases. The proposed neural network model optimized by ROA has greater gains as compared to other algorithms like a Hippopotamus Optimization (HO) and Puma Optimization (PO) algorithms.