Privacy-Preserving Data Mining Process in Industry

Protecting the privacy of users in web-scale data mining applications and systems like online searching, recommender systems, crowd-sourced platforms, and analytics apps has developed into an incredibly essential component in recent years. This is especially true in light of recent data breaches and new legislation such as GDPR. In this tutorial, we'll go over a history of privacy violations from the last two decades, along with the lessons we gained from them, as well as important rules and legislation and how privacy tactics have evolved into different privacy definitions and techniques. Data mining raises unique privacy risks, despite being successful in many applications. The literature has looked at the trade-offs that the current system presents between utility and personal privacy. However, industrial deployment has its own unique challenges when it comes to the management of resource trade-offs between computation and communication. The industrial deployments that are taking place right now are rather scattered. Each device (such a mobile phone, for example) is responsible for storing a single data point, and it is the server's job to compute aggregates based on the data that is obtained from the distinct mix of public and private information that is associated with each device. In spite of the fact that computing on servers is quite inexpensive, computing on clients is somewhat pricey since clients are often comprised of low-power devices. The expectation maximization (EM) technique for distribution reconstruction is discussed in this study. It is superior to the existing approach regarding the degree of information loss. The percentage of information loss is only 4%. In particular, we show that the EM approach, when applied to data that has been subjected to a perturbation, eventually arrives at the maximum likelihood estimate of the distribution that was being modeled. We demonstrate that the EM algorithm produces reliable estimations of the original distribution when considerable data is available.