ENHANCING BIG DATA PRIVACY AND PERFORMANCE THROUGH EDGE-INTEGRATED FEDERATED LEARNING: A COMPARATIVE STUDY

Abstract

The study provides a new model to combine federated learning (FL) with edge computing, differential privacy (DP), and homomorphic encryption (HE) to improve privacy preservation and performance in working with big data. The model was tested on synthetic datasets that resemble soft copies of the Aadhaar card. The proposed model was compared with traditional FL, DP, and HE models. In this paper, we presented a hybrid privacy-preserving model that generates synthetic Aadhaar records, resulting in 100,000 records. And compared the performance of various privacy-preserving methods---Federated Learning, Differential Privacy, Homomorphic Encryption, and hybrid model---using execution time, memory utilization, CPU usage, and privacy preservation, as evaluation metrics. The experimental results demonstrate that the proposed model offers the strongest privacy preservation performance under the defined metric, along with competitive execution time (0.03 seconds), indicating a favorable trade-off between privacy and efficiency. The proposed method fits the needs of data and identity verification processes that are sensitive and need to be secure.