Demystifying fractional order chaotic respiratory disease system with XAI

Donmez, Turker Berk, Kutlu, Mustafa, Freeman, Chris, Akgul, Akif and Karaca, Yeliz (2024) Demystifying fractional order chaotic respiratory disease system with XAI. In The 3rd International Conference on Applied Mathematics in Engineering (ICAME’24). p. 138 .

Abstract

The current study delves into the intricate association between meteorological conditions and the incidence of Upper Respiratory Tract Infections (URTIs), leveraging the advanced capabilities of the CatBoost machine learning algorithm in conjunction with a Fractional Order Chaotic System and cutting-edge Explainable Artificial Intelligence (XAI) techniques. By analyzing comprehensive meteorological and health data collected from the Pamukova District (Marmara Region, Turkey), this research paper employs the SHapley Additive exPlanations (SHAP) values to elucidate the model’s predictions, emphasizing the consequential effects of mean temperature, humidity, and atmospheric pressure over a 5-day period on the occurrence of URTIs [1, 2]. The findings obtained by the related analyses demonstrate that mean temperature holds a dominant influence on URTI predictions, with SHAP values peaking at 5.6, thus underscoring its critical role as a predictive marker for increased URTI cases. Similarly, the mean humidity is identified as a pivotal factor, manifesting a maximum SHAP value of 3.2, which signifies its substantial impact on the prevalence of URTIs. In contrast, mean pressure exhibits a wide array of SHAP values, indicating a multifaceted and somewhat indirect correlation with URTI incidences [3].
Integral to our approach is the incorporation of a fractional-order system that meticulously accounts for the history of data, thereby offering a nuanced understanding of the temporal dynamics influencing URTI trends. This aspect of our methodology not only enriches the predictive model with a deeper temporal context but also aligns with the foundational principles of chaotic systems as described by
Lorenz, enhancing the robustness and accuracy of our predictions [1].
The predictive prowess of our model is evidenced by an accuracy rate of 75.21%, complemented by precision and recall metrics of 0.75 and 0.5217, respectively. Such metrics highlight the feasibility and effectiveness of our integrated approach in forecasting URTI occurrences with considerable reliability.
The implications of our study are far-reaching for the domain of public health, accentuating the imperative to integrate extended weather data within disease prediction frameworks and to inform efficient and timely targeted preventive measures and strategies.

This record has no associated files available for download.

Contributors

Author: Chris Freeman

Download statistics