Predicting COVID-19 using Bayesian Network

This study details a study aimed at improving the diagnosis of Coronavirus disease (COVID-19) and Long COVID using Bayesian Networks. Leveraging comprehensive data collection from online sources, the research endeavoured to replicate and enhance a previously conducted study that employed Bayesian Networks as they can handle missing data, are based on probabilistic reasoning and have transparency and interpretability, for the same purpose. 

Rigorous data cleaning and pre-processing stages were implemented to ensure the reliability of the data. The Bayesian Network model was constructed and refined, incorporating prior probabilities via the Bayesian estimator and maximum likelihood estimation method. Adherence to best practices in machine learning, including stringent cross-validation and sensitivity analysis, was maintained throughout the model implementation. Though the diagnosis of COVID-19 based solely on symptoms, conditions, and demographic factors remains a significant challenge, the research underlines the potential of Bayesian networks as a valuable tool in tackling complex predictive tasks.

The model developed has a 93% chance of correctly distinguishing between patients with and without COVID-19. Similarly, the PR curve is 0.88, which reflects the precision and recall balance of our Bayesian model. For long COVID, we have an AUC of 0.55 on the ROC curve and 0.41 on the PR curve suggesting that the model’s ability to differentiate between patients with and without Long COVID is not significantly better than flipping a coin.

The study underscores the effectiveness of BN in predicting COVID-19 and highlights the need for further refinement in Long COVID prediction