CNN-Based Classification of Cardiovascular Disease: A Hyperparameter Optimization and Dimensionality Reduction Approach

This work presents a comprehensive two-stage model that integrates structured clinical datasets and electrocardiogram (ECG) data to predict Cardiovascular disease (CVD) with greater precision. The first stage employs classical ML methods, utilizing a Random Forest algorithm on merged and pruned clinical datasets, including Framingham and Statlog, to derive initial predictions. The second stage refines these predictions using a Convolutional Neural Network (CNN) trained on ECG image data, enabling deeper insights and improved classification performance. The model was developed on the Jupyter Notebook platform using Python, leveraging libraries such as TensorFlow, Keras, Scikit-learn, and Pandas. Preprocessing involves converting ECG images into 1D signals, followed by dimensionality reduction. Techniques like Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) are used for dimensionality reduction. These techniques streamlined the feature set, retaining only the most discriminative information and enhancing model efficiency. Hyperparameter optimization for the CNN was performed using GridSearchCV, fine-tuning key parameters such as filters, activation functions, batch size, and epochs to maximize performance. By integrating clinical and ECG data through a multi-layered approach, this study demonstrates the potential for an advanced, highly accurate model for CVD prediction. The results underline the effectiveness of combining traditional ML techniques with DL-based analysis for holistic and robust healthcare solutions.