Analysis of Machine Learning Model with Differential Anonymization Technique
Amardeep Singh, Monika Singh · 2025
Concern over user privacy is growing in today's data-driven society. In this paper, the effects of anonymization approaches have been analysed on publicly available dataset that contains sensitive information that may be vulnerable to re-identification attempts. The dataset of diabetes from UCI repository has been anonymized and tested with machine learning model - neural networks using differential privacy with Laplace noise. The results show that the model performs with an accuracy of 96.1% when data is anonymized using our technique of differential privacy with minimum loss of utility of the dataset.