Introduction: Pediatric Gastroenteritis—A Persistent Global Challenge

Pediatric gastroenteritis remains a major public health concern worldwide, particularly in low- and middle-income countries, where it contributes significantly to morbidity and mortality among children under five years of age. Despite improvements in sanitation, nutrition, and vaccination, the burden of disease remains high, with seasonal outbreaks placing additional stress on healthcare systems. These outbreaks often follow predictable seasonal patterns—such as rotavirus infections peaking in winter and bacterial gastroenteritis surging during summer—yet they are also shaped by multifactorial and interrelated risk factors including age, nutritional status, hygiene practices, vaccination coverage, and environmental conditions.

Traditional epidemiological models often struggle to account for the complex interactions among these variables and their temporal fluctuations. As such, there is a growing need for more sophisticated predictive tools capable of analyzing heterogeneous datasets and recognizing nonlinear trends. Artificial intelligence (AI), particularly deep learning, offers promising solutions for predictive epidemiology by integrating diverse data sources to enhance accuracy in disease forecasting.

The VGG-Dense HybridNetClassifier is a newly developed deep learning model designed to meet these challenges. By combining the spatial feature extraction capabilities of VGG networks with the information-rich connectivity of DenseNet architectures, this hybrid model excels at capturing both static and dynamic patterns associated with pediatric gastroenteritis. Recent studies demonstrate its capacity to predict disease risk across time and population segments with greater accuracy than traditional models, offering a valuable framework for early warning systems, resource allocation, and targeted public health interventions.

Methods: Architecture of the VGG-Dense HybridNetClassifier

The VGG-Dense HybridNetClassifier is a dual-branch neural network architecture that integrates convolutional and dense layers to analyze both structured and unstructured data relevant to pediatric gastroenteritis. The convolutional branch is based on a VGG architecture—typically VGG-16 or VGG-19—chosen for its efficiency in hierarchical feature extraction from time-series data, image inputs, or encoded representations of temporal trends, such as seasonality or symptom progression.

In parallel, the dense branch processes structured tabular inputs such as electronic health record (EHR) data, demographic profiles, weather indices, and environmental health indicators. This dual-stream approach allows the model to learn complementary features from heterogeneous modalities: spatial and temporal trends from one stream, and static or categorical risk factors from the other.

The outputs from both branches are fused in a concatenated layer, followed by fully connected layers that refine the joint representation before classification. The model is trained to predict gastroenteritis risk using multicenter datasets representing diverse pediatric cohorts across geographical and socioeconomic contexts. Hyperparameter optimization, including adjustment of learning rate, dropout rate, and batch size, is guided by cross-validation strategies such as k-fold and stratified sampling to ensure robustness and generalizability.

Compared to single-architecture models, the VGG-Dense HybridNetClassifier has demonstrated stronger capabilities in learning from noisy, imbalanced, or temporally complex datasets, positioning it as a promising tool for precision public health in pediatric populations.

Results: Model Performance and Key Findings

Evaluation of the VGG-Dense HybridNetClassifier across multiple datasets revealed excellent predictive accuracy, with overall performance metrics exceeding those of standard machine learning baselines. The model achieved an accuracy of over 92% and an AUC-ROC greater than 0.95, while also delivering balanced precision, recall, and F1-scores. These metrics indicate reliable identification of both high-risk and low-risk cases across different seasons and population subgroups.

Importantly, the model effectively captured subtle, nonlinear interactions among variables such as age, geographic location, pathogen type, vaccination status, and meteorological conditions. This enabled not only precise early prediction of individual risk but also robust stratification of seasonal transmission patterns. For instance, the model successfully predicted rotavirus surges in cooler months and enteric bacterial outbreaks during warmer, more humid conditions—patterns consistent with known epidemiological data.

Interpretability was enhanced through visualization techniques such as heatmaps and attention maps, which highlighted the relative influence of features such as precipitation, temperature, age under two years, and incomplete immunization. These insights contribute to the model’s transparency and usability in clinical or public health decision-making.

Validation across independent, multicenter cohorts confirmed the model’s generalizability and clinical relevance, demonstrating consistent performance across different patient populations and environmental contexts. By identifying early warning signals and facilitating proactive responses, the VGG-Dense HybridNetClassifier provides actionable, data-driven forecasts that support targeted interventions, improved surveillance, and optimized resource deployment during pediatric gastroenteritis outbreaks.

Reference:

  1. Al-Banna S, Al-Banna N, Al-Banna I. Prevalence of Acute Gastroenteritis Enteropathogens Among Children in Jordan: A Cross-Sectional Study. Pathogens. 2025 Apr;14(4):500.

  2. World Health Organization. Diarrhoeal disease fact sheet. Updated March 2024.

  3. Kirk MD, Pires SM, Black RE, et al. Global burden of diarrheal diseases and risk factors in children under five. Lancet Infect Dis. 2024;24(3):e123-e134.

  4. Al-Muharraqi MA, Al-Haddad AM, Al-Khalifa MS. Seasonal variation and molecular epidemiology of rotavirus and norovirus in pediatric gastroenteritis. J Med Virol.2023;95(7):e28241.

  5. Pranesh, P.T., Singh, C.D., Sivanandam, A. et al. Predictive analysis of pediatric gastroenteritis risk factors and seasonal variations using VGG Dense HybridNetClassifier a novel deep learning approach. Sci Rep 15, 23912 (2025). https://doi.org/10.1038/s41598-025-08718-4

  6. Paul S, Rahman M, Dolley A, Saikia K, Shyamsunder Singh C, Mohammed A, Muteeb G, Sarmah R, Namsa ND. A retrospective study using machine learning to develop predictive model to identify rotavirus-associated acute gastroenteritis in children. PeerJ. 2025 Apr 14;13:e19025. doi: 10.7717/peerj.19025. PMID: 40247842; PMCID: PMC12005185.

  7. Qian YF, Guo WL. Development and validation of a deep learning algorithm for prediction of pediatric recurrent intussusception in ultrasound images and radiographs. BMC Med Imaging. 2025 Mar 3;25(1):67. doi: 10.1186/s12880-025-01582-8. PMID: 40033220; PMCID: PMC11874663.

  8. Ogwel B, Mzazi VH, Nyawanda BO, Otieno G, Tickell KD, Omore R. A machine learning approach to predicting inpatient mortality among pediatric acute gastroenteritis patients in Kenya. Learn Health Syst. 2024 Dec 26;9(2):e10478. doi: 10.1002/lrh2.10478. PMID: 40247897; PMCID: PMC12000769.

  9. Zhao Q, Gao Q, Guo X, Han Y, Zhang J, Yang Y, Tang J, Shi J, He L, Xiong T. Machine learning-based diagnostic model for neonatal intestinal diseases in multiple centres: a cross-sectional study protocol. BMJ Open. 2025 May 14;15(5):e096750. doi: 10.1136/bmjopen-2024-096750. PMID: 40374225; PMCID: PMC12083289.