ADAGRAD-ENHANCED DEEP LEARNING MODELS FOR HEART DISEASE CLASSIFICATION

Abstract

Heart disease classification is a critical area in medical diagnostics, requiring accurate and efficient methodologies. This study explores the application of the Adagrad optimizer to enhance the performance of three deep learning models—Multilayer Perceptron with DNN (MLP-DNN), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM)—in classifying heart disease. This work pre-process the dataset from kaggle which containing clinical feature. Also the data set undergo feature engineering to produce engineered attributes suitable for deep learning models, this research work conduct extensive experiments encompassing data pre-processing, feature selection, and Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance. These research findings demonstrate the effectiveness of Adagrad in optimizing model performance across all architectures. Furthermore, detailed analysis reveals insights into the strengths and limitations of each model in classifying different disease subtypes, highlighting their applicability in clinical contexts. While the MLP-DNN model exhibits overall superiority, variations in performance across specific disease categories underscore the importance of tailored model selection based on diagnostic requirements. This study contributes to advancing heart disease classification by introducing an innovative optimization strategy that enhances model convergence and accuracy. Future research avenues may explore hybrid architectures, ensemble techniques, and larger datasets to further enhance classification performance and facilitate real-world deployment in clinical settings.