PRIVACY PRESERVING SEQUENTIAL PATTERN MINING OVER DISTRIBUTED PROGRESSIVE DATABASES

Abstract

Frequent Sequential Pattern Mining, commonly known as Sequential Pattern Mining is a data mining technique used to find interesting patterns in a large collection of data items. A common example of sequential pattern mining from market-basket databases is to track customer-buying patterns between the different items purchased by them. The discovery of such patterns can help retailers develop marketing strategies by gaining insight into frequently purchased items and their trends. The databases used for these purposes are progressive databases, which are a generalized model providing dynamic addition and deletion of data for efficient mining operations. Sometimes a group of local market players may be interested in mining trends by pooling in their individual data. However the shared data may disclose some information which might be against the privacy policies of these collaborating parties or may be of strategic importance for some party. The need for a privacy preserving mechanism is thus felt to safeguard the sensitive information shared during the mining process. In our dissertation work, we propose a set of algorithms for finding sequential patterns from distributed databases while preserving privacy. The work aims at maintaining the privacy of the data and patterns mined with minimal effect on accuracy of the results. In this work, the algorithms address all three types of fragmentation (viz. Vertical, Horizontal, Arbitrary).The proposed work of sequential pattern mining is applicable to progressive databases(special cases being static and incremental databases). In this work we use cryptographic and randomization techniques to achieve privacy preservation. The work also proposes an idea to suppress sensitive sequential pattern mining results. This proposition has generally been applied to the various kinds of distributed databases under study.