S.B.K. U., Mahalaxmi, I, Kala, Kumar Sagar, Lalit, M, Thomas, Kaushal, Ashish Kumar and Rio Robert, Nismon (2025) Transfer learning techniques and approaches for predictive modeling of disease outcomes. Informing Science: The International Journal of an Emerging Transdiscipline, 28. ISSN 1547-9684
![[thumbnail of InfoSciV28Art016Mahalaxmi11109.pdf]](https://pure.jgu.edu.in/style/images/fileicons/text.png)
InfoSciV28Art016Mahalaxmi11109.pdf - Published Version
Download (591kB)
Abstract
Aim/Purpose
In this research work, we have developed a predictive model that focuses on utilizing knowledge from the related domains.
Background
A serious public health issue, especially in tropical and subtropical regions, is dengue fever, a viral infection passed by mosquitoes. Accurate early prediction of disease outcomes is essential for both efficient patient management and effective use of resources. More complex methods are required since conventional prediction models could be faulty with limited labeled data and complex feature interactions.
Methodology
We propose a new strategy integrating deep attention mechanisms with transfer learning to enhance prediction modeling of dengue disease outcomes. First pre-trained on a large, linked dataset of common viral illnesses, a deep neural network enables the model to learn generic properties. We then iteratively improve our pre-trained model using a specific dengue dataset. Incorporating a deep attention mechanism allows for the focus on the most relevant features, improving interpretability and accuracy.
Contribution
Among logistic regression, random forests, and basic deep learning methods, current models reveal poor accuracy and dependability in forecasting dengue disease outcomes. These models sometimes fail to sufficiently depict the complicated interactions among clinical variables, especially under conditions with limited data.
Findings
The proposed method outperforms more traditional models pretty strongly. Our model acquired in the training phase an accuracy of 0.92, precision of 0.91, recall of 0.90, and F1-score of 0.90. It maintained high performance on testing with an accuracy of 0.91, precision of 0.90, recall of 0.89, and an F1-score of 0.89. Similar patterns were indicated by an accuracy of 0.90, precision of 0.89, recall of 0.88, and an F1-score of 0.88 validation results. The model also demonstrated a lowered loss (0.21, 0.23, 0.24 in training, testing, and validation, respectively), higher true positive rates (0.90, 0.89, 0.88), and lower false positive rates (0.10, 0.11, 0.12). Deep attention methods and transfer learning offer a robust and effective strategy for predictive modeling of dengue disease outcomes, therefore considerably boosting accuracy and dependability. This approach offers considerable possibilities for dengue-endemic patient management and resource allocation.
Recommendation for Researchers
Investigations should prioritize the validation of the algorithm in various healthcare environments to assess its efficacy in clinical application.
Future Research
In future research, this work can be enhanced using several deep learning algorithms to achieve better accuracy and performance.
Item Type: | Article |
---|---|
Keywords: | dengue prediction | transfer learning | deep attention mechanisms | predictive analytics | healthcare informatics |
Subjects: | Physical, Life and Health Sciences > Public Health, Environmental and Occupational Health Social Sciences and humanities > Social Sciences > Social Sciences (General) |
JGU School/Centre: | Jindal Global Business School |
Depositing User: | Dharmveer Modi |
Date Deposited: | 01 Apr 2025 13:30 |
Last Modified: | 01 Apr 2025 13:30 |
Official URL: | https://doi.org/10.28945/5443 |
URI: | https://pure.jgu.edu.in/id/eprint/9336 |
Downloads
Downloads per month over past year