Abstract
This article discusses a neural network ensemble algorithm for clinical decision support tasks, using diabetes prediction as an example. Using an ensemble of models, including convolutional neural networks (CNNs), recurrent neural networks (RNNs), and long short-term memory (LSTM) networks, improves the accuracy and robustness of predictions compared to individual architectures. The Dirichlet Ensemble method, based on the Dirichlet distribution, is used to combine predictions. This method provides flexible and adaptive weighting of results, taking into account the confidence in each model and data uncertainty. Experiments included a comparative analysis of the models' performance, in which RNN demonstrated superior performance in terms of accuracy, recall, and F1 score. The ensemble approach improved the reliability and robustness of predictions, which is critical for clinical decision making and minimizing diagnostic errors. The study confirms the relevance of neural network ensembles and probabilistic fusion methods in medicine, contributing to the development of digital decision support systems. The practical significance of this work lies in the potential for integrating the proposed algorithm into real-world medical information systems to improve the diagnostic quality of diabetes and other chronic diseases. Further research prospects include expanding the ensemble architecture and optimizing the weights using adaptive methods.