Table 1 A summary of the related works with advantages and disadvantages.

Altantawy et al.²⁴

Deep attentive model

Faisalabad

Accuracy = 99.2%

(i) It effectively handles high dimensional data (ii) Better Handling of Imbalanced Data

(i) It has high computational complexity (ii) Susceptible to over-fitting problem

CVD

Accuracy = 92.7%

Heart failure

Accuracy = 97%

Jafar et al.²⁵

HyperOpt optimizer-LASSO optimizer

Cleveland

Accuracy = 97.32%

(i) The model effectively handles the linear and non-linear data (i) It increases the Prediction performance

(i) It produces poor performance on noisy data (ii) It is an inefficient algorithm, it takes a longer time to process

CVD

Accuracy = 97.72%

Omkari et al.²⁶

Integrated TLV framework

UCI

Accuracy = 99.03%

(i) A maximum accuracy is achieved (ii) Improved Handling of Imbalanced Data

(i) Increase computational overhead (ii) Needs a proper feature selection algorithm (iii) High model complexity

CVD

Accuracy = 88.09%

Mandava²⁷

IDRSNet

UCI

Specificity = 98.95%

Sensitivity = 98.90%

Accuracy = 99.12%

(i) It predicts heart disease with higher accuracy (ii) The probability of over-fitting is decreased

(i) It has a higher amount of missing values (ii) It requires an efficient feature selection technique (iii) It has poor data generalization

Tata et al.²⁸

Deep VAE AEO

Framingham

Accuracy = 97%

Precision = 98%

Recall = 87%

F1-score = 82%

(i) Efficient handling of imbalanced data (ii) Enhanced feature extraction with VAE (iii) Robustness to noisy and incomplete data

(i) Parameter tuning can be complex and time-consuming (ii) High model complexity

Nandakumar et al.²⁹

Inception-resNet-V2

UCI Cleveland

Accuracy = 98.77% Precision = 87%

F1-score = 90%

Specificity = 85%

Sensitivity = 93%

(i) Fast convergence and solving local optima problems (ii) It produces an improved accuracy (iii) Highly efficient for handling noisy data

(i) The training and testing run times are roughly longer than for other models. (ii) Need for regularization techniques (iii) Sensitivity to Class Imbalance

Revathi et al.³⁰

OCI-LSTM

UCI

Accuracy = 97.11%

Precision = 98%

Recall = 87%

F1-score = 82%

(i) Insensitive to irrelevant features (ii) Manages both continuous and discontinuous data

(i) It requires significant resources for training and tuning. (ii) High dimensionality feature space and uneven sample sizes for the target classes.

Elsedimy et al.³¹

QPSO-SVM

Cleveland

Accuracy = 96.31%

Precision = 94.23%

Recall = 96.13%

F1-score = 95%

ii) It produces an improved accuracy ii) Highly efficient for handling noisy data

i) The training and testing run times are roughly longer than for other models. (ii)It requires significant resources for training and tuning

Torthi et al.³²

BAPSO-RF

UCI

Accuracy = 98.71%

Precision = 98.67%

Recall = 98.23%

F1-score = 98.45%

(i) Ensures high classification stability and robustness (ii) Suitable for large-scale biomedical datasets

(i) Particle swarm variants may get trapped in local optima.

(ii) Computationally expensive when tuning both BAPSO and RF parameters

Kumar et al.³³

CapsNet-B-KHA

Cleveland

Accuracy = 95%

Precision = 94%

Recall = 97%

F1-score = 95%

i) Captures spatial hierarchies in feature representation. (ii) Better generalization and interpretability via capsule structures.

(i) High training time and memory usage compared to CNNs.

(ii) Capsule networks are still relatively new—lack of framework maturity.

Kumar et al.³⁴

The sample-based neural network

CVD

Accuracy = 96%

Precision = 97%

Recall = 95%

F1-score = 95%

i) Capable of handling imbalanced datasets via sample reweighting.

(ii) Simplifies network complexity for small datasets.

(iii) Reduces over-fitting with fewer trainable parameters.

i) May suffer from lower performance on large datasets. (ii) Sensitive to sample selection strategies.

Arunachalam et al.³⁵

Ensemble model

UCI

Accuracy = 96%

Precision = 97%

Recall = 95%

F1-score = 95%

(i) Aggregates multiple weak learners to improve prediction accuracy.

(ii) High robustness to noisy data.

(iii) Effective in handling high-dimensional feature spaces.

(i) Increased model complexity and interpretability issues. (ii) Requires more training time and computation resources.

Saranya et al.³⁶

DenseNet-ABiLSTM

ECG signal data

Accuracy = 89.14%

F1-score = 87.74%

(i) DenseNet provides effective feature reuse and gradient flow. (ii) ABLSSTM enhances temporal sequence understanding in ECG signals. (iii) Strong performance in real-time sequential prediction tasks.

(i) High memory requirements for DenseNet layers. (ii) Difficult to optimize due to multiple deep components.