Abstract
Globally, cardiovascular diseases (CVDs) remain an important cause of mortality, for early and precise diagnosis. Traditional machine learning (ML) models struggle with feature redundancy, data imbalance, and suboptimal performance due to inefficient feature selection and lack of robust deep learning (DL) architectures. To overcome these challenges, we propose a novel Cardiovascular Disease Stage Classification via the Mamdani Fuzzy-based Modified Mobile Network (CVD-M 3 Net) by integrating hybrid feature selection, DL network and Fuzzy logic for improved CVD stage classification. The proposed CVD-M 3 Net utilizes data from multiple sources including Cleveland, MIMIC-IV, UK Biobank, Statlog, Switzerland, Hungary, and VA Long Beach datasets. The Boruta-LASSO and Fast ICA techniques are used for feature selection by ensuring the retention of critical diagnostic attributes while eliminating irrelevant ones. The selected features undergo Z-score normalization for improved data consistency between the 15 features. The Modified MobileNet (MOMO-Net) is introduced with the integration of 1D convolutional and transformer layers in the MobileNet structure to categorize the tabular data for real-time CVD stage detection. The proposed CVD-M 3 Net utilizes the Mamdani fuzzy inference system (FIS) with a trapezoidal membership function to predict the CVD Stage Score (CSS). The efficiency of the proposed CVD-M 3 Net was estimated with the Accuracy, Sensitivity, Precision, Recall, and F1-score. From the experimental analysis, the proposed CVD-M 3 Net achieves an overall accuracy of 98.61% for efficient classification of CVD stages. The proposed CVD-M 3 Net increases the accuracy by 0.11%, 0.62%, 3.25%, and 0.38% better than ML algorithms, O-SBGC-LSTM, MaLCaDD, and DL-based CNN, respectively.