الفهرس 
Related WorkOnly 14 pages are availabe for public view
 |  | from | 137 |  |  |
| | | from | 137 | | |
Abstract
Abstract
Over the last two decades, due to the broad usage of the internet and information technologies, there has been an enormous increase in the use of data mining applications for data processing and analysis. However, the advancement of data mining technologies has posed a severe danger to the privacy of individuals’ sensitive information and raised several ethical issues. These privacy issues lead to major obstacles to information exchange and also prevent users from providing their crucial information. As a reaction to this problem, Privacy Preserving Data Mining (PPDM) algorithms have been proposed. Privacy Preserving Utility Mining (PPUM) is a crucial and popular topic in the field of PPDM, where it is an integration of both Utility Pattern Mining (UPM) and PPDM. The main aim of this thesis is to propose an effective PPDM framework to improve data sharing and publishing possibilities. The proposed framework includes two modules: high utility pattern mining module and privacy preserving utility mining module.
The first module proposes an effective detection approach for the top-k high utility patterns (HUPs) in transactional datasets. The proposed approach adopts novel techniques to efficiently find top-k HUPs. Finding these patterns can greatly enhance the decision-making processes in many applications and various fields, especially in the field of e-commerce and market engineering. The experimental results proved that the proposed top-k HUPM approach outperforms the state-of-the-arts in terms of execution time (up to 10,000 faster) and memory consumption (up to 87% lower memory).
The second module of our proposed framework handles the security issues of the utility mining techniques. Applying UPM techniques may be of great benefit, but some of the resultant HUPs may also reveal sensitive knowledge about the data owner. Three novel PPUM algorithms are proposed in this module, which allows achieving not only the privacy and security requirements but also guarantees valid data mining results. Experiments show promising results in terms of reducing the sanitization time (up to 90% lower) and negative effects (60% lower) compared to the state-of-the-arts.
List of Publications