An Interpretable Machine Learning Framework For Early Prediction Of Febrile Seizure Risk In Children

Mohammed Hussein Ghazi; Muhammed Sabri Salim

doi:10.29284/p5akbw51

Authors

Mohammed Hussein Ghazi Department of Electronics & Communication Engineering, College of Engineering, Al-Nahrain University, Baghdad, Iraq.
Muhammed Sabri Salim Department of Electronics & Communication Engineering, College of Engineering, Al-Nahrain University, Baghdad, Iraq.

DOI:

https://doi.org/10.29284/p5akbw51

Keywords:

Febrile Seizures, Embedded Healthcare Systems, Pediatric Risk Prediction, Clinical Decision Support Systems.

Abstract

Background: Febrile seizures are the most frequent cause of convulsive disorder during childhood and still represent a great discomfort for the patients' families and also for the treating physician. Although most cases have a benign course, clinical value of finding the children with higher risk in advance is nonetheless favorable for low resources or real-time monitoring.

Objective: This research work was designed to create a predictive model of early onset of febrile seizures with clinically interpretable predictions, low computational cost, and with the strict focus on embedded-system implementation and the resulting data visualization to gain more clinical understanding.

Methods: Monte Carlo simulation was used to obtain a synthetic dataset of 1,000 cases of pediatrics subject to aggregated epidemiologic statistics of a published clinical study. Predictor variables were age, sex, body temperature, family history of febrile seizures and maternal smoking during pregnancy. A stratified 70/30 train test split was used to train a logistic regression model. The accuracy of classification, receiver operating characteristic (ROC) analysis, and odds ratios with corresponding confidence intervals were used to assess model performance. The last model took the form of a closed-form mathematical equation that can be employed using low-power embedded systems.

Results: The proposed model had a classification accuracy of 83.67% and area under ROC curve (AUC) of around 0.90. Family history of febrile seizures was found to be the most powerful predictor (OR > 8), then high body temperature and maternal smoking in pregnancy. Visual inspection indicated that there were clear distributions patterns in the seizure and non-seizure groups in all the predictors. The resulting equation was numerically stable and of low computational complexity and can be implemented in real-time on resource-constrained machines.

Conclusions: Clinically meaningful prediction of Febrile seizure prediction can be done with a small number of variables easily available and with clinically meaningful prediction. The suggested framework is feasible as it offers a viable base of interpretable, entrenched clinical choice-support frameworks and presents a scalable path toward validation of the same with genuine clinical data in the future. The visual analytics help to improve clinical knowledge of risk factor distributions and interactions.

References

[1] D. Laino, E. Mencaroni, S. J. I. j. o. e. r. Esposito, and p. health, "Management of pediatric febrile seizures," vol. 15, no. 10, p. 2232, 2018.

[2] S. Shinnar and T. A. J. J. o. c. n. Glauser, "Febrile seizures," vol. 17, no. 1_suppl, pp. S44-S52, 2002.

[3] V. Thadchanamoorthy and K. J. I. N. D. J. Dayasiri, "Review on febrile seizures in children," vol. 12, pp. 25-35, 2020.

[4] A. T. Berg and S. J. E. Shinnar, "Complex febrile seizures," vol. 37, no. 2, pp. 126-133, 1996.

[5] A. J. E. j. o. p. Fetveit, "Assessment of febrile seizures in children," vol. 167, no. 1, pp. 17-27, 2008.

[6] A. K. Leung, K. L. Hon, and T. N. J. D. i. c. Leung, "Febrile seizures: an overview," vol. 7, p. 212536, 2018.

[7] D. C. Hesdorffer et al., "Distribution of febrile seizure duration and associations with development," vol. 70, no. 1, pp. 93-100, 2011.

[8] A. Ferretti et al., "Best practices for the management of febrile seizures in children," vol. 50, no. 1, p. 95, 2024.

[9] E. J. J. N. m. Topol, "High-performance medicine: the convergence of human and artificial intelligence," vol. 25, no. 1, pp. 44-56, 2019.

[10] J. Kubota et al., "Body temperature predicts recurrent febrile seizures in the same febrile illness," vol. 43, no. 7, pp. 768-774, 2021.

[11] M. L. Castellazzi, A. La Vecchia, M. Scali, C. Agostoni, G. Di Pietro, and G. P. J. F. i. P. Milani, "Clinical and laboratory parameters associated with febrile seizure recurrence within the first 24 h: a ten-year cohort study," vol. 12, p. 1373848, 2024.

[12] F. Shen et al., "Risk factors and predictors of recurrence of febrile seizures in children in Nantong, China: a retrospective cohort study," vol. 24, no. 1, p. 420, 2024.

[13] G. Yalçın et al., "A Comprehensive Perspective on Febrile Seizures in Children: A Prospective Cohort Study with Evaluation of Clinical, Laboratory, and Genetic Features," vol. 14, no. 22, p. 7918, 2025.

[14] J. Papež et al., "Multivariate linear mixture models for the prediction of febrile seizure risk and recurrence: a prospective case–control study," vol. 13, no. 1, p. 17372, 2023.

[15] A. Cheng et al., "Development and validation of a predictive model for febrile seizures," vol. 13, no. 1, p. 18779, 2023.

[16] M. S. Salim, N. Sabri, A. A. R. J. E. A. S. Dheyab, and Technology, "Predicting WSN packet loss using machine learning: Applications in solid surroundings," vol. 9, no. 2, pp. 289-302, 2025.

[17] F. Doshi-Velez and B. J. a. p. a. Kim, "Towards a rigorous science of interpretable machine learning," 2017.

[18] C. J. N. m. i. Rudin, "Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead," vol. 1, no. 5, pp. 206-215, 2019.

[19] M. Halimeh, M. Jackson, S. Vieluf, T. Loddenkemper, and C. J. S. E. J. o. E. Meisel, "Explainable AI for wearable seizure logging: Impact of data quality, patient age, and antiseizure medication on performance," vol. 110, pp. 99-108, 2023.

[20] S. Ramgopal, L. N. Sanchez-Pinto, C. M. Horvat, M. S. Carroll, Y. Luo, and T. A. J. P. r. Florin, "Artificial intelligence-based clinical decision support in pediatrics," vol. 93, no. 2, pp. 334-341, 2023.

[21] M. Dosani et al., "Clinical evaluation of algorithms for context-sensitive physiological monitoring in children," vol. 102, no. 5, pp. 686-691, 2009.

[22] L. Forsgren, R. Sidenvall, H. K. S. BLOMQUIST, J. Heijbel, and L. J. A. P. Nyström, "Pre‐and perinatal factors in febrile convulsions," vol. 80, no. 2, pp. 218-225, 1991.

[23] Z. Zou, B. Chen, D. Xiao, F. Tang, and X. J. J. o. M. I. R. Li, "Accuracy of Machine Learning in Detecting Pediatric Epileptic Seizures: Systematic Review and Meta-Analysis," vol. 26, p. e55986, 2024.

[24] G. Yücel, A. K. Arslan, B. Özgör, and S. J. B. P. O. Güngör, "Prediction of recurrent febrile seizures risk during the same febrile illness in children at a single tertiary centre in Turkiye," vol. 9, no. 1, p. e002908, 2025.