Securing Data Privacy in Machine Learning: The FedAvg of Federated Learning Approach
Shiva Mehta, Aseem Aneja
Abstract
This research evaluates the FedAvg method that is created for picture categorization, particularly the important aspects of privacy and data security in the machine learning field. The research employs a dataset that has 60.000 training photos and 10.000 test images with ten labels that correspond to different fashion items classifying. The FedAvg technique trains a CNN model by negotiating model updates in several virtual parties where a peer-to-peer server collects the model updates from each of the parties and agglomerates them to optimize the CNN model. The results show high achievement, due to this fact trained model has gained an accuracy of 87.27% on the test dataset. the result of statistics is one indicator of system efficiency since the data is close to the ones of the centralized training approaches. This document, therefore, brings to the limelight the possibility of federated learning as a privacy-preserving method that conceivably manifests both data privacy and general accuracy in picture classification assignments. This technique has the chance to be developed and used in different areas where the confidentiality of data is a matter that has to be considered. The dataset consisted of 60,000 training shots and 10,000 test images, categorized into ten classes: Cotton t-shirt, cotton pair of pants, cotton dress, cotton sweater, cotton coat, cotton sandals, cotton shirt, leather shoes, cotton bag, and cotton boots. After every step, the image showed up as a black and white image with dimensions of 28 by 28. The study demonstrates the positive effect of Federated Averaging on image classification. It reveals the effectiveness of this method and gives results similar to the ones obtained by the centralized training approaches that have been in use until now. This research works out the practical convenience within which Federated learning operates by delivering an accuracy of 87.27%. Therefore, there will be more compliance with the provision of tight data privacy where regulations must be followed.