Analyzing Female Consumers’ Adoption of Online Grocery Platforms in India: A Technology Acceptance Model (TAM) Perspective

Abstract

Machine learning (ML) models are widely used in various domains, such as healthcare, finance, and social media. However, ML models may also pose privacy risks, as they can reveal sensitive information about the training data or the users who interact with them. To protect the privacy of data and users, several techniques have been proposed, such as anonymization and differential privacy. Anonymization aims to remove or modify the identifying attributes of the data, such as names, addresses, or phone numbers. Differential privacy adds random noise to the data or the model outputs, such that the presence or absence of any individual in the data does not affect the results significantly. However, both techniques have limitations and challenges, such as information loss, utility degradation, or high computational cost. In this paper, we propose a novel hybrid algorithm that combines anonymization and differential privacy to enhance the security of data and ML models. Our algorithm applies k-anonymity with n-gram to the data before sending it for training with ML model, which further processed with differential privacy. Differential privacy allows performing computations on encrypted data without decrypting it, while blowfish encryption is a fast and secure symmetric-key algorithm. Our algorithm ensures that the data and the model are protected from unauthorized access or modification by the malicious third parties. We evaluate our algorithm on several benchmark datasets and ML models, and show that it achieves high accuracy and privacy while reducing the communication and computation overhead. We also compare our algorithm with existing methods and demonstrate its advantages and limitations.