Table 1.
Comparison of existing model in literature.
| Key technique | Model | Research performance | Limitation | References |
|---|---|---|---|---|
| Machine learning and deep learning | CNN, LSTM, KNN, XGB, and majority voting ensemble | Proposed model obtained the highest classification performance based on all evaluation metrics on all datasets | Potential limitations in generalizability to other populations or datasets, need for further validation, May require significant computational resources | 30 |
| Deep learning | CNN-GRU, SMOTE Method | Higher classification accuracy compared to other existing models | Potential limitations in generalizability to other datasets or environments | 31 |
| Ensemble learning, data mining techniques | Weighted ensemble model using genetic algorithm | Improved performance compared to individual classifiers | May require significant computational resources | 32 |
| Remote monitoring | Web application for remote monitoring and management, Real-time monitoring and alerts | Effective in monitoring and managing high-risk pregnancies | Limited to healthcare professionals, not designed for patient use | 32 |
| Ensemble-based deep learning model | CNN, LSTM, XGBoost, KNN | Outperformed existing models, demonstrating superiority in cardiovascular disease prediction | Lack of interpretability of the model’s predictions due to the complexity of the ensemble architecture | 33 |
| Semantic relatedness and similarity measures | natural language, machine learning algorithms | Using students’ answers as feedback considerably improved the accuracy and performance of these measures | The dataset used is relatively small | 34 |
| Machine learning | Neural networks, SVM, KNN | remarkable accuracy and minimal loss | Limited to a single dataset, potential variation with other datasets | 35 |
| Machine learning | Nomogram prediction model | Successfully identified several parameters associated with stroke risk, demonstrated superior predictive accuracy | Potential limitations in generalizability to other populations, need for further validation | 36 |
| Machine learning (ML) | Random forest (RF), KNN, DT, AdaBoost, XGBoost, SVM, ANN | RF achieved highest performance | Potential limitations in generalizability to other populations or datasets, need for further validation, May require significant computational resources | 37 |
| Ensemble Machine Learning | Soft Voting Classifier (Random Forest, Extremely Randomized Trees, Histogram-Based Gradient Boosting) | Achieved an accuracy of 96.88%, improved accuracy and robustness compared to single classifiers | Potential limitations in handling complex interactions between features, need for further optimization | 18 |
| Face Detection using Yolo v8 | Stroke monitoring strategy | Achieved high accuracy of 98.43% | Limited availability of stroke patient data | 38 |
| Modified Vision Transformer (ViT) integrated approach | End to end ViT Architecture, CNN | 87.51% classification accuracy for brain CT scan slices | Improvement needed for stroke diagnosis | 43 |
| A deep-learning-based Microwave-induced thermo acoustic tomography MITAT (DL-MITAT) Technique | A residual attention U-Net (ResAttU-Net) | effectively eliminated image artifacts and accurately restored hemorrhage spots as small as 3 mm | No performance metrics for increased accuracy; training sets are constructed only using the simulation approach | 44 |
| AutoML | A combination of AutoML, Vision Transformers (ViT), and CNN | The model achieved 87% accuracy for single-slice level predictions and 92% accuracy for patient-wise predictions | Small sample size, complexity of the integrated architecture | 45 |