Probabilistic Decision Tree Ensembles for Detecting Infectious Diseases using Routine Blood Tests

The use of new machine learning methods for detecting pandemics has achieved significant attention in recent years. Involving the development and the evaluation of various learning approaches on different datasets, a research effort that has confirmed its effectiveness in explaining decision-making and classification processes. Particularly in the detection of infectious diseases using routine blood tests. In this study, the quality of decision-making is investigated by proposing a fusion of two decision tree algorithms, Random Forest and XGBoost, applied to routine blood test data from patients tested for Covid-19 admitted to hospital in Italy. The combination is achieved using the theory of belief functions, especially Dempster-Shafer theory, to improve classifications previously made by these two learning algorithms. Based on the concepts of uncertainty, an attenuation of masses has also been applied to the probabilities of the least performing algorithm. The attenuation parameter used was defined using genetic algorithm for the optimization of the proposed Probabilistic Ensemble Learning Model (PELM). The combination achieved an accuracy rate of 92.31 %, a precision rate of 92.30%, a recall rate of 92.31 %, and an F1-score of 92.28%.