Optimization of Transformer heart disease prediction model based on particle swarm optimization algorithm

Jingyuan Yi; Peiyang Yu; Tianyi Huang; Zeqiu Xu

REVIEW 4 major objections 7 minor 4 cited by

Optimization of Transformer heart disease prediction model based on particle swarm optimization algorithm

T0 review · 4 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that using particle swarm optimization to tune a Transformer's hyperparameters achieves 96.5% classification accuracy on a heart disease dataset, beating random forest by 4.3 percentage points.

desk verdict PSO-tuned Transformer for heart disease data reports 96.5% accuracy, but missing validation split, absent Transformer baseline, and sloppy citations leave the central claim unsupported. read the letter →

arxiv 2412.02801 v3 pith:7MFHPYI5 submitted 2024-12-03 cs.AI

Jingyuan Yi , Peiyang Yu , Tianyi Huang , Zeqiu Xu This is my paper · ORCID

classification cs.AI

keywords heartdiseasepredictionparticleswarmoptimizationTransformerhyperparameterclassificationrandomforestmedicalmachinelearning

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to show that particle swarm optimization (PSO) can tune a Transformer model to classify heart disease more accurately than standard machine learning baselines. On an open-source dataset of 1888 patient records, the PSO-optimized Transformer reportedly reaches 96.5% accuracy, versus 92.2% for random forest, 91.2% for decision tree, and 80.8% for XGBoost. If this holds, it would mean that a biologically inspired search algorithm can replace manual hyperparameter selection for Transformer models on medical tabular data. The result matters because accurate heart disease prediction could support earlier intervention and better use of health care resources.

What carries the argument

The central object is the PSO-Transformer hybrid: particle swarm optimization treats hyperparameter combinations as particles moving through a multidimensional search space, with velocity updated from each particle's own best position and the swarm's global best, and fitness measured by model accuracy or loss during training. The Transformer itself is the base model, using multi-head self-attention and positional encoding to make predictions from tabular patient features. This machinery does the work of replacing manual hyperparameter search with an automated, swarm-driven search over model depth, width, learning rate, and attention head count.

What would settle it

Train the same PSO-Transformer pipeline with an explicit three-way split — training, validation for PSO fitness, and a held-out test set touched only once — and compare the final test accuracy with random forest on the same split. If the independently measured accuracy falls to roughly the random forest level or below, the paper's central claim of a PSO-driven improvement would be refuted; if it stays near 96.5%, the claim is supported.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that PSO can optimize the architecture and training hyperparameters of a Transformer — specifically learning rate, number of layers, hidden layer dimension, and number of attention heads — so that the resulting model outperforms leading classical classifiers on the same heart disease prediction task. The claim is supported by a single comparison: after 100 PSO iterations with 30 particles, the optimized Transformer is reported to achieve 96.5% accuracy, 92.6% precision, 97.3% recall, and 96.2% F1 score, compared with 92.2% accuracy for random forest, the best of the three baselines. The paper concludes that the 4.3 percentage point improvement verifies the effectiveness of PSO in optimizing Transformer models for this task.

Load-bearing premise

The central claim assumes that the accuracy used to score each PSO particle's fitness came from a validation set that was never used to select the final model; if the same test set that produced the 96.5% figure was used for fitness evaluation, the reported advantage could be an artifact of overfitting.

Editorial extensions

If this is right

PSO can automate hyperparameter selection for Transformers, replacing manual tuning in this setting.
The optimized Transformer outperforms decision tree, random forest, and XGBoost on the same dataset, indicating that attention-based models with tuned hyperparameters are competitive on tabular medical data.
Because PSO also optimizes structural parameters such as layer count and attention heads, the method could be applied to other prediction tasks without task-specific architecture engineering.
The reported 4.3 percentage point gain over random forest is the concrete evidence for the paper's conclusion that PSO effectively optimizes Transformer performance.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The paper evaluates only one dataset of 1888 records; whether the 4.3 point gain generalizes to other heart disease cohorts or to non-medical tabular datasets is an open question the paper does not address.
Because no validation split is described, a safer reading is that the headline accuracy may include selection on the test set; a nested cross-validation would clarify whether PSO's gain is real or an artifact.
An implicit comparison the paper does not make is against a PSO-tuned random forest or a default Transformer; without those, the improvement cannot be attributed specifically to PSO versus the Transformer architecture.
The PSO settings are fixed at 30 particles, 100 iterations, and inertia weight 0.9; testing sensitivity to these settings would show how robust the reported 96.5% accuracy is.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

This paper applies standard PSO to Transformer hyperparameter search on a public heart-disease dataset and reports 96.5% accuracy, about 4.3 points above random forest. The problem is that the experiment is missing the controls needed to support that headline. There is no validation split described anywhere, and the PSO fitness function is said to use accuracy or loss. If the same test set that produced the 96.5% was used to evaluate particles, that number is a selection artifact, not an independent measurement. Even if the split were clean, the paper never trains a Transformer without PSO, so the comparison to random forest cannot isolate the contribution of PSO. The gap could be due to the Transformer architecture itself or to chance.

To be fair, the paper does some things competently. It gives a legible walk-through of PSO and Transformer basics, uses a standard public dataset, reports confusion matrices, and lists hyperparameters. That is more than some arXiv submissions do. But the positives stop there. There is no cross-validation, no error bars, no code, and no dataset link. The references are messy: Vaswani is cited for the dataset and again for PSO with autoencoders, Kazemnejad's blog on positional encoding is cited for PSO+SVM, and there is a placeholder "Doe et al." entry. The claim that PSO "verifies" effectiveness is not justified.

This looks like a routine application of a known optimizer to a small tabular problem, with nothing new in method or theory. The global-health framing in the abstract and conclusion is far out of proportion to the incremental result. I don't think this paper is important or solid enough to warrant referee time. A serious editor should desk reject it, though the authors might benefit from feedback on experimental design and citation hygiene.

For a reader wanting a simple example of PSO for Transformer tuning, this might illustrate the general idea, but the missing controls make the specific result unreliable. I would not cite it, and I would not bring it to a reading group.

Referee Report

4 major / 7 minor

Summary. The paper proposes using particle swarm optimization (PSO) to tune hyperparameters of a Transformer model for heart disease prediction. The authors first train decision tree, random forest, and XGBoost models on an open-source dataset described as containing 1888 samples, reporting random forest as the best baseline with 92.2% accuracy. They then train a Transformer whose learning rate, number of layers, hidden dimension, and attention heads are optimized by PSO, and report 96.5% accuracy, concluding that PSO significantly improves Transformer performance. The paper includes no code, no dataset identifier, no train/validation/test split description, no non-PSO Transformer baseline, and no error bars or significance testing.

Significance. If the reported claim were fully supported, the paper would provide a modest but potentially useful demonstration that PSO-based hyperparameter optimization can improve Transformer accuracy on a small medical tabular dataset. However, the central claim is currently unverified because the experimental design lacks the two essential controls: a Transformer trained without PSO under identical conditions, and a clear separation between the data used for PSO fitness evaluation and the test set used for the reported accuracy. The paper also provides no reproducibility artifacts (code, data, or precise experimental protocol). Given these omissions, the paper in its current form does not establish that PSO contributed to the reported performance, and the contribution reduces to an accuracy number without scientific support.

major comments (4)

[Section 4.4, Table 2] The central claim that PSO improves Transformer performance is unsupported because the paper never trains a Transformer without PSO. The comparisons in Table 2 are against decision tree, random forest, and XGBoost, which are different model classes; any accuracy gap between the PSO-Transformer and these baselines could be due to the Transformer architecture alone, the PSO tuning, or both. An experiment with a Transformer trained using default or randomly chosen hyperparameters, evaluated on the same data split and with the same protocol, is necessary to isolate the effect of PSO. Without this control, the 96.5% accuracy does not demonstrate that PSO contributed anything.
[Section 4.4] The manuscript never describes a validation set. The text states that PSO evaluates each particle's fitness on 'specific tasks (e.g., accuracy, loss, etc.)' but does not specify whether the fitness accuracy is computed on a validation split, the test set, or the training set. If the same test set used for the final reported 96.5% in Table 2 is also used to compute PSO fitness, then the reported accuracy is a selection artifact caused by optimizing hyperparameters on the test set, not an independent estimate of generalization. The authors must explicitly define the train/validation/test split and confirm that PSO used only the validation split.
[Section 2 and Section 4.4] The experimental protocol lacks cross-validation, repeated runs, and error bars. With a dataset of 1888 samples and a single reported accuracy per model, the 4.3 percentage point gap between the PSO-Transformer (0.965) and random forest (0.922) could be within run-to-run variability, especially for a stochastic PSO search and neural network training. The authors should report mean and standard deviation over multiple independent runs, or use k-fold cross-validation, and ideally perform a statistical significance test (e.g., McNemar's test) to support the claim of improvement.
[Section 2, Table 1] The dataset is not adequately described or correctly cited. The abstract says the target variable is coded 0 (no heart disease) and 1 (heart disease), but Section 2 says '1 in target indicates yes, and 2 indicates no', and Table 1 shows only rows with target equal to 1. The dataset is attributed to 'Vaswani et al. [2017a]', which is the 'Attention is all you need' Transformer paper, not a heart disease dataset. The authors must identify the exact dataset (e.g., the UCI Heart Disease dataset or a specific variant), report the number of classes, class balance, feature preprocessing, and how the 1888 samples were split into training, validation, and test sets. Without this information, the experimental results are not reproducible.

minor comments (7)

[Table 2] The row label 'Textual model' appears to be a typo and should be 'Transformer model'.
[Section 4.1] The claims that PSO with stacked sparse auto-encoders achieves 96.1% accuracy and that PSO has been used to optimize SVM are cited to Vaswani et al. [2017b] and Kazemnejad [2020], respectively; neither reference is related to those claims. Appropriate references for PSO-SAE and PSO-SVM should be provided.
[Section 2] The phrase 'totaling 1888 pieces of information' should be '1888 samples' or '1888 patient records' for clarity.
[Section 4.2] The PSO algorithm is described verbally but the velocity and position update equations are not given. Adding the standard PSO update equations would improve precision and reproducibility.
[Section 4.4] The model parameter section lists fixed values for learning rate (0.001), hidden dimension (512), attention heads (8), and feedforward dimension (2048), but does not state whether these are the values found by PSO or the settings used for a non-PSO baseline. The authors should clarify the parameter search ranges and report the optimized hyperparameter values.
[Section 3 and Figure 9] The correlation analysis in Section 3 reports no quantitative results, and Figure 9's caption 'The comparison of model parameters' is vague; specify what is being compared (e.g., model size, training time, or hyperparameter values).
[Abstract and Conclusion] The conclusion states that PSO 'significantly improves' Transformer performance, but no statistical significance testing is reported; the wording should be tempered to match the evidence.

Circularity Check

1 steps flagged · score 6.0 of 10

PSO fitness is the same accuracy metric reported as the headline result, with no validation split described, so the 96.5% figure is a fitted objective rather than an independent test.

fitted input called prediction [Section 4.4 (Improved Transformer model based on PSO), Table 2]
"These combinations are then used to train the Transformer model and evaluate its performance on specific tasks (e.g., accuracy, loss, etc.) to determine fitness values. The particle then updates its speed and position to explore more optimized hyperparameters. Through many iterations, the PSO can find the optimal combination of hyperparameters."

PSO's fitness function is stated to be model performance measured by accuracy or loss, and the paper's only reported evaluation metric is accuracy. The final claim, 'the classification accuracy of the model is as high as 96.5%,' uses the same quantity that PSO is described as maximizing. The manuscript never describes a train/validation/test split or any holdout used for final evaluation, so on the paper's own description the reported 96.5% is the optimized fitness value, not an independent prediction. This is the fitted-input-called-prediction pattern: the 'result' is the objective being optimized, so it cannot by itself verify that PSO improved the Transformer.

full rationale

The paper's derivation chain is minimal: compare three classical ML models to a PSO-optimized Transformer and report accuracy. The central circularity concern is that the same accuracy metric serves both as the PSO fitness function and as the headline outcome, and no data split is documented. If a held-out test set had been described and used only for final reporting, the result would be legitimate; as written, the 0.965 figure is indistinguishable from the optimized objective. This is partial circularity because the improvement is claimed relative to random forest, which is at least an external benchmark, but the claim that PSO specifically improved the Transformer is not independently testable from the paper: there is no non-PSO Transformer baseline, no optimized hyperparameter values reported, no code, and no data split. However, the missing baseline is a completeness problem, not circularity. There are no self-citations, and no uniqueness theorem or ansatz is imported from the authors' prior work, so the score is dominated by the fitness/accuracy reduction.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the hand-chosen PSO and Transformer hyperparameters, an unknown data split, and the unstated assumption that the fitness metric used by PSO is independent of the test set. No new entities are introduced.

free parameters (3)

PSO hyperparameters (particles, iterations, inertia, c1, c2) = 30 particles, 100 iterations, w=0.9, c1=c2=2
Hand-chosen values; no sensitivity analysis or justification is provided.
Transformer architecture (learning rate, layers, hidden dim, heads, feedforward dim) = lr=0.001, batch=32, dim=512, heads=8, ff=2048
Some values are set manually and some are optimized by PSO; the paper does not separate which values come from where.
Train/test split = not reported
The dataset split is never described, so all reported metrics depend on an unknown split.

assumptions (3)

standard math Pearson correlation is an appropriate measure for assessing feature-target relationships.
Section 3 uses Pearson correlation without discussing non-linear relationships or the categorical nature of several features such as sex and chest pain type.
domain assumption The 1888-record open-source dataset is representative of the target population.
Section 2 gives no dataset citation or characteristics, and the described columns resemble the UCI Cleveland dataset, which has far fewer records.
ad hoc to paper PSO fitness measured as accuracy is a valid proxy for generalization.
Section 4.4 uses accuracy or loss as the fitness function, but no separate validation set is described, so the proxy may be the test set itself.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Optimization of Transformer heart disease prediction model based on particle swarm optimization algorithm." pith.science (2026). https://pith.science/paper/7MFHPYI5

@misc{pith2026241202801,
  author       = {Pith},
  title        = {Pith review of: Optimization of Transformer heart disease prediction model based on particle swarm optimization algorithm},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7MFHPYI5}},
  note         = {Machine review of arXiv:2412.02801}
}

read the original abstract

Aiming at the latest particle swarm optimization algorithm, this paper proposes an improved Transformer model to improve the accuracy of heart disease prediction and provide a new algorithm idea. We first use three mainstream machine learning classification algorithms - decision tree, random forest and XGBoost, and then output the confusion matrix of these three models. The results showed that the random forest model had the best performance in predicting the classification of heart disease, with an accuracy of 92.2%. Then, we apply the Transformer model based on particle swarm optimization (PSO) algorithm to the same dataset for classification experiment. The results show that the classification accuracy of the model is as high as 96.5%, 4.3 percentage points higher than that of random forest, which verifies the effectiveness of PSO in optimizing Transformer model. From the above research, we can see that particle swarm optimization significantly improves Transformer performance in heart disease prediction. Improving the ability to predict heart disease is a global priority with benefits for all humankind. Accurate prediction can enhance public health, optimize medical resources, and reduce healthcare costs, leading to healthier populations and more productive societies worldwide. This advancement paves the way for more efficient health management and supports the foundation of a healthier, more resilient global community.

Figures

Figures reproduced from arXiv: 2412.02801 by the authors.

**Figure 1.** Correlation heat maps. For the analysis of this correlation heat map, we can see that the values in the figure range from -1 to 1, indicating the strength of the correlation between different variables. The color depth of the thermal map can intuitively indicate the intensity of the correlation. The darker the color, the 2 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** The schematic diagram of particle swarm opti [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** The schematic diagram of Transformer. Transformer architecture is mainly divided into Encoder and Decoder two parts. The function of the encoder is to encode the input sequence into a representation vector that contains contextual information. Each encoder layer consists of two main components: a self-attention mechanism and a fully connected feedforward neural network. The self-attention mechanism allows the model… view at source ↗

Figures from the paper (5 more)

**Figure 4.** Figure 4: The optimization process. In the experiment of Transformer classification algorithm based on particle swarm optimization (PSO) algorithm, the hardware configuration is 3060 graphics card and the memory is 32G. 4 [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

**Figure 7.** Figure 7: The confusion matrix of theXGBoost model. [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]

**Figure 8.** Figure 8: The confusion matrix of our model [PITH_FULL_IMAGE:figures/full_fig_p005_8.png]

**Figure 5.** Figure 5: The confusion matrix of the decision tree model. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

**Figure 9.** Figure 9: The comparison of model parameters. fication and prediction of heart disease, with an accuracy rate of 92.2%; while the classification accuracy rate of Transformer model optimized based on particle swarm optimization in this paper reaches 96.5%, 4.3% higher than that …

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Mixed-Precision Graph Neural Quantization for Low Bit Large Language Models
cs.CL 2025-01 conditional novelty 6.0 of 10

A graph neural network trained on Hessian-derived graphs of LLM weights adaptively assigns per-column quantization bit-widths, improving low-bit perplexity over GPTQ on LLaMA models.
Machine Learning Algorithm for Noise Reduction and Disease-Causing Gene Feature Extraction in Gene Sequencing Data
cs.LG 2025-05 reject novelty 4.0 of 10

A CNN+RNN denoiser and an ensemble classifier are claimed to improve sequencing noise reduction and disease-gene prediction, but the paper provides no code, data, or reproducible evidence.
Mitigating Knowledge Conflicts in Language Model-Driven Question Answering
cs.CL 2024-11 reject novelty 4.0 of 10

On memorized question-answer pairs from KMIR and NQ, bottleneck and prefix adapters trained on entity-swapped contexts let a GPT-2 reader follow the new context most of the time, though no baselines are reported.
A Hybrid Transformer Model for Fake News Detection: Leveraging Bayesian Optimization and Bidirectional Recurrent Unit
cs.CL 2025-02 reject novelty 2.0 of 10

Adding a vaguely specified Bayesian component to a BiGRU-Transformer raises reported fake news test accuracy from 99.67% to 99.73% on one Kaggle dataset, with no code, data, or error bars.

Reference graph

Works this paper leans on

12 extracted references · 11 canonical work pages · cited by 4 Pith papers

[1]

Vijay Arya, Rachel K. E. Bellamy, Pin-Yu Chen, Amit Dhurandhar, Michael Hind, Samuel C. Hoffman, Stephanie Houde, Q. Vera Liao, Ronny Luss, Aleksandra Mojsilovi\' c , Sami Mourad, Pablo Pedemonte, Ramya Raghavendra, John Richards, Prasanna Sattigeri, Karthikeyan Shanmugam, Moninder Singh, Kush R. Varshney, Dennis Wei, and Yunfeng Zhang. Ai explainability ...

work page arXiv 2021
[2]

Heart disease diagnosis using deep learning

John Doe, Alice Smith, Richard Brown, and Michael Lee. Heart disease diagnosis using deep learning. In IEEE Conference on Biomedical Systems, volume 7, pages 120--128. IEEE, 2024

work page 2024
[3]

Particle swarm optimisation: A historical review up to the current developments

Diogo Freitas, Luiz Guerreiro Lopes, and Fernando Morgado-Dias. Particle swarm optimisation: A historical review up to the current developments. Entropy, 22 0 (3): 0 362, 2020. doi:10.3390/e22030362

work page doi:10.3390/e22030362 2020
[4]

A comparison of particle swarm optimization and the genetic algorithm

Rania Hassan, Babak Cohanim, Olivier de Weck, and Gerhard Venter. A comparison of particle swarm optimization and the genetic algorithm. MIT Engineering Systems Division Working Paper Series, pages 1--30, 2004. URL http://dspace.mit.edu/handle/1721.1/6018

work page 2004
[5]

Sahadat Hossain, Md

Md. Sahadat Hossain, Md. Alamin Talukder, and Md. Zulfiker Mahmud. Advancements in cardiovascular disease detection: Leveraging data mining and machine learning. bioRxiv, 2024. doi:10.1101/2024.03.09.584222

work page doi:10.1101/2024.03.09.584222 2024
[6]

Hussein and A

R. Hussein and A. Abdulazeez. Skin cancer detection utilizing deep learning: Classification of skin lesion images using a vision transformer. Journal of Applied Science and Technology Trends, 5 0 (2): 0 60--71, 2024. Discusses deep learning algorithms like CNNs and their applications in distinguishing between benign and malignant skin lesions

work page 2024
[7]

Understanding self-attention and positional encoding of the transformer architecture

Amirhossein Kazemnejad. Understanding self-attention and positional encoding of the transformer architecture. Personal Blog, 2020. URL https://kazemnejad.com/blog/transformer_architecture_positional_encoding/

work page 2020
[8]

Global cardiovascular disease statistics and impacts: Key findings and challenges

World Health Organization. Global cardiovascular disease statistics and impacts: Key findings and challenges. World Health Statistics Report, 2024. Available at https://www.who.int

work page 2024

Show all 12 references

[9]

A secure remote health monitoring for heart disease prediction using machine learning and deep learning techniques in explainable artificial intelligence framework

Sibo Prasad Patro and Neelamadhab Padhy. A secure remote health monitoring for heart disease prediction using machine learning and deep learning techniques in explainable artificial intelligence framework. In Proceedings of the 10th International Electronic Conference on Senso...

2023 doi
[10]

Particle swarm optimization: A survey of historical and recent developments with hybridization perspectives

Saptarshi Sengupta, Sanchita Basak, and Richard Alan II Peters. Particle swarm optimization: A survey of historical and recent developments with hybridization perspectives. Machine Learning and Knowledge Extraction, 1 0 (1): 0 157--191, 2019. doi:10.3390/make1010010

2019 doi
[11]

Gomez, Łukasz Kaiser, and Illia Polosukhin

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in Neural Information Processing Systems, 30: 0 5998--6008, 2017 a

2017
[12]

Attention is all you need

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in Neural Information Processing Systems, 30: 0 5998--6008, 2017 b

2017

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

Vijay Arya, Rachel K. E. Bellamy, Pin-Yu Chen, Amit Dhurandhar, Michael Hind, Samuel C. Hoffman, Stephanie Houde, Q. Vera Liao, Ronny Luss, Aleksandra Mojsilovi\' c , Sami Mourad, Pablo Pedemonte, Ramya Raghavendra, John Richards, Prasanna Sattigeri, Karthikeyan Shanmugam, Moninder Singh, Kush R. Varshney, Dennis Wei, and Yunfeng Zhang. Ai explainability ...

work page arXiv 2021

[2] [2]

Heart disease diagnosis using deep learning

John Doe, Alice Smith, Richard Brown, and Michael Lee. Heart disease diagnosis using deep learning. In IEEE Conference on Biomedical Systems, volume 7, pages 120--128. IEEE, 2024

work page 2024

[3] [3]

Particle swarm optimisation: A historical review up to the current developments

Diogo Freitas, Luiz Guerreiro Lopes, and Fernando Morgado-Dias. Particle swarm optimisation: A historical review up to the current developments. Entropy, 22 0 (3): 0 362, 2020. doi:10.3390/e22030362

work page doi:10.3390/e22030362 2020

[4] [4]

A comparison of particle swarm optimization and the genetic algorithm

Rania Hassan, Babak Cohanim, Olivier de Weck, and Gerhard Venter. A comparison of particle swarm optimization and the genetic algorithm. MIT Engineering Systems Division Working Paper Series, pages 1--30, 2004. URL http://dspace.mit.edu/handle/1721.1/6018

work page 2004

[5] [5]

Sahadat Hossain, Md

Md. Sahadat Hossain, Md. Alamin Talukder, and Md. Zulfiker Mahmud. Advancements in cardiovascular disease detection: Leveraging data mining and machine learning. bioRxiv, 2024. doi:10.1101/2024.03.09.584222

work page doi:10.1101/2024.03.09.584222 2024

[6] [6]

Hussein and A

R. Hussein and A. Abdulazeez. Skin cancer detection utilizing deep learning: Classification of skin lesion images using a vision transformer. Journal of Applied Science and Technology Trends, 5 0 (2): 0 60--71, 2024. Discusses deep learning algorithms like CNNs and their applications in distinguishing between benign and malignant skin lesions

work page 2024

[7] [7]

Understanding self-attention and positional encoding of the transformer architecture

Amirhossein Kazemnejad. Understanding self-attention and positional encoding of the transformer architecture. Personal Blog, 2020. URL https://kazemnejad.com/blog/transformer_architecture_positional_encoding/

work page 2020

[8] [8]

Global cardiovascular disease statistics and impacts: Key findings and challenges

World Health Organization. Global cardiovascular disease statistics and impacts: Key findings and challenges. World Health Statistics Report, 2024. Available at https://www.who.int

work page 2024