Privacy-preserving machine learning tool for mitigating data leakage in microservices architectures

Abstract

In the era of distributed systems and microservices architectures, the risk of data leakage, particularly of personally identifiable information (PII), has become a critical concern. Federated Learning (FL) emerges as a promising solution by enabling collaborative model training across decentralized data sources without the need to transfer raw data to a central server, thereby preserving user privacy. This study implemented a privacy-preserving federated learning utilizing Flower frammework, an open-source FL platform, in conjunction with TensorFlow for model development. The Federated Averaging (FedAvg) algorithm was employed as the core aggregation strategy to combine model updates from multiple clients. A hybrid deep learning model was designed to optimize learning across the distributed network, ensuring both robustness and scalability. Over the course of five training rounds, two clients participated in each round, contributing locally trained model weights to a central aggregator. The performance of the federated model was rigorously evaluated using standard machine learning metrics: accuracy, loss, F1-score, Area Under the ROC Curve (AUC), and precision. The results demonstrated progressive improvement in model performance, with accuracy increasing from 76.0% in round 1 to 87.7% in round 5, and AUC improving from 0.88 to 0.93, indicating enhanced classification capability over time. Similarly, F1-score and precision showed consistent growth, signifying improved balance between precision and recall and reduced false positives. The distributed training also showcased a decreasing loss trend, dropping from 1.018 to 0.891 across the rounds, reflecting better model convergence. Importantly, this study illustrated the practical viability of federated learning for privacy-centric applications, showing that high-performance machine learning models can be achieved without compromising data privacy. Future work can focus on scaling this approach to a larger number of clients, integrating differential privacy and secure aggregation techniques to further strengthen privacy guarantees, and comparing the hybrid model with alternative architectures to enhance model generalizability across heterogeneous data sources.

Keywords: Federated Learning, Privacy Preservation, Flower Framework, TensorFlow, Federated Averaging (FedAvg), Hybrid Deep Learning Model, Machine Learning Metrics, Aggregation