A Comparative Review of Privacy-Preserving Federated Learning Algorithms for Allergy and Asthma Severity Assessment

Abstract

The application of artificial intelligence (AI) for allergy and asthma severity assessment involves the processing of highly sensitive patient data, raising significant concerns related to privacy, security, and regulatory compliance [1]. Federated learning (FL) is a decentralized learning approach that allows multiple healthcare entities to jointly train models without exchanging raw patient data.[2]. This paper presents a comparative review of privacy-preserving federated learning algorithms, including Federated Averaging (FedAvg) [2] and Federated Proximal (Fed Prox) [3], in conjunction with classical and advanced machine learning models such as logistic regression, random forest, gradient boosting, and Long Short-Term Memory (LSTM) networks. The reviewed algorithms are evaluated based on key performance metrics, including predictive accuracy, privacy protection, computational complexity, and scalability in multi-institutional healthcare environments. In addition, privacy- enhancing techniques such as secure aggregation [4], differential privacy [5], and homomorphic encryption [6] are examined for their effectiveness in mitigating data leakage while maintaining model performance. The analysis indicates that hybrid frameworks integrating federated learning with ensemble or deep learning models and robust privacy mechanisms provide an optimal balance between model accuracy and data confidentiality. These findings offer valuable insights for the development of secure, scalable, and high-performance AI systems for real-world allergy and asthma severity assessment in distributed healthcare networks.