Federated learning-based uroflowmetry pattern classification and home health services integrated decision support model

Over time, interest in artificial intelligence (AI) and machine learning (ML) in the healthcare sector has been growing exponentially. Considering that there are relatively few studies on the use of artificial intelligence technology in the field of urology, applications in this field will have a significant impact on decision making. Big data is needed to develop machine learning models. However, as healthcare organizations strive to collect clinical records on centralized servers to develop robust machine learning models, concerns about privacy, data ownership and legal constraints arise. Federated learning (FL) approach is one of the promising technologies that allows models to evolve across multiple data centers without the need to share sensitive information. In the federated learning process, model training is distributed and each device trains a local model using its own data. The parameters from the local models are then aggregated and a global model is constructed. In this study, uroflowmetry curve patterns of pediatric patients were classified using uroflowmetry test data obtained from a urology outpatient clinic. A federated learning system, centralized and local learning approaches are compared. An artificial neural network architecture with optimized hyperparameters using the Archimedes Optimization Algorithm (AOA) metaheuristic was used to train the models. According to the findings, federated learning algorithms exhibited a classification performance very close to centralized learning and were found to be superior to client-based local learning. The Model-Contrastive Federated Learning (MOON) algorithm achieved 84% classification accuracy. The prediction outputs obtained in the study were used as input in a bi-objective scheduling problem for home health services (HHS), and the mathematical model developed in this direction was successfully solved in a short time using a commercial solver. The intelligent decision support model presented in this thesis offers an important innovation by contributing to data-driven decision making processes in the field of urology.