Enhancing explainability in epidemiological predictions using fuzzy logic integrated with machine and deep learning algorithms

Epidemiological data is often analyzed without fully accounting for the uncertainties that are key to understanding the nuances of the dataset. While traditional approaches like the SIR mathematical model provide valuable insights, our study aims to offer a different perspective by employing fuzzy logic. This methodology assigns weightage to selected factors, creating a framework that not only captures uncertainties but also improves data handling and interpretation. By reducing the number of features, our approach makes datasets more manageable while preserving their essential information. We applied this method to two epidemiological datasets, the H1N1 and seasonal vaccine dataset and the COVID-19 dataset. Initially, classical machine learning algorithms SVM, ensemble algorithm XGBoost, and deep learning algorithm ANN, were used to analyze the data. Subsequently, the features were fuzzified, resulting in the development of fuzzy machine learning and fuzzy deep learning algorithms. These algorithms consistently delivered reliable results within acceptable ranges. To further validate their effectiveness, we tested the proposed algorithms on datasets from other domains, including a diabetes healthcare dataset and a student performance dataset. The results demonstrated enhanced insights, optimized outcomes, and improved data management. Our study highlights how fuzzy logic can provide better insights and address uncertainties, offering a novel approach to studying epidemiological data.