Machine Learning-Based Data Analytics With Privacy: Privacy-Preserving Data Analytics

Background

Privacy preservation of collected data is important in respect of decision making of analyzed data. This is true in every domain like healthcare, IoT based systems, business intelligence, decision support system etc. In literature data perturbation is done using randomization, K-anonymity, data suppression and data generalization. This is applicable to all types of data as stated in the work (Ge et al., 2005). Machine learning, deep learning, distributed machine learning are areas to work on privacy (Zhou et al., 2021; Niu et al., 2020; Mohassel & Zhang, 2017). K-anonymity works on the principle of hiding individuality of data. But it becomes difficult while dealing with high dimensional data as it becomes difficult to preserve privacy (Shokri & Shmatikova, 2015; Sweeney, 2002). Datafly, µ-Argus and k-Similar are some k-anonymities based modified algorithms presented in literature (Aggarwal, 2005). Some group-based privacy preserving algorithms are also proposed which works on different privacy aspects (Majid Rafiei, 2021). Author discussed methods of privacy preservation, metrics to assess privacy and application areas of privacy preservation in detail in (Mendes & Vilela, 2017). A review of differential privacy preserving in machine learning for balancing privacy and utility of data is put forth by author (Gong et al., 2020).