HIDING SENSITIVE PATTERNS FOR PRIVACY PRESERVATION

Abstract

Frequent pattern mining is an area of interest of data mining and is used for identifying strong association rules or patterns from large databases. Frequent pattern mining is often used across organizations, which collaborate with each other in order to get mutual benefits. The results that are obtained after performing frequent pattern mining on the joint databases of the collaborative parties provides the global results which are helpful in making effective business strategic plans and decisions. To get the global mining results, the collaborative parties need to shares their databases. Sharing of database with others can leads to privacy threat to individual parties because the database may contain some sensitive knowledge. This sensitive knowledge may contain some business trending information or private information that the owner party does not want to disclose to other parties. So before publishing the database to other parties, it is required to obtain a privacy preserved database so that sensitive knowledge cannot be mined from the new modified database. Frequent itemset or association rule hiding is a well-known technique for masking sensitive patterns. Hiding of sensitive frequent itemsets also have side effects on the original database, these side effects include masking of some useful non-sensitive frequent itemsets and generation of new artificial patterns which are not present in the original database. So the aim of hiding algorithms is to not only hide sensitive patterns but also minimize the side effects on the original database. In sensitive pattern hiding different approaches are used such as heuristic based, border based etc. It has been observed that the heuristic based approaches provide fast and easy solution but these approaches have more side effects on original database whereas border based approaches have less side effects as compared to heuristic based approaches. To overcome the limitations of the heuristic based approaches, we utilize the border-based approach due to the fact that it has fewer side effects on the original database and proposed an IBHA (Improved Border-based Hiding Algorithm) algorithm. The results of the proposed IBHA algorithm has been compared with existing heuristic and border based algorithms and the proposed method gives much better results in the form of lesser side effects.