Data transformation approaches for privacy preserving data mining

Recent advances in data mining techniques facilitate to explore hidden knowledge from a large volume of data. When organizations share data for mining, they may restrict confidential information and knowledge to the other organizations. To protect sensitive information before data sharing, the modern age of information processing has evolved a new research area, namely Privacy Preserving Data Mining. Data transformation methods facilitate to preserve privacy without losing the benefit of data mining. The existing studies have dealt with data transformation methods for numerical data to preserve privacy in clustering and also data sanitization approaches to hide sensitive patterns. It is essential to devise new data transformation methods for categorical data to preserve privacy in clustering. The existing data sanitization approaches are capable of removing a number of legitimate patterns while concealing sensitive patterns. They also focus exclusively on specific pattern types. Nevertheless, it is necessary to develop new data sanitization approaches to hide sensitive patterns. In this work, to begin with, sensitive categorical data protection in clustering is addressed. Two hybrid data transformation methods have been devised to transform the sensitive categorical data. Then, their effectiveness in privacy preservation and clustering accuracy are validated. It is found that iv scaling and rotation transformation method improves the privacy level and the translation and rotation transformation method provides better accuracy in clustering. Hiding sensitive association rules are implemented by concealing the frequent itemsets. It includes the concepts of non-sensitive item conflict degree, item and transaction conflict ratio. Experimental results indicate that the use of item and transaction conflict ratio reduces the legitimate itemsets missed after sanitization. The work further focuses on sanitization approaches for privacy preservation of sensitive utility itemsets. With an intention to deal with this, two data sanitization approaches are devised using transaction conflict degree and item conflict degree. The experimental results indicate that the item conflict degree improves results in terms of the legitimate itemsets lost. Privacy preservation of utility and frequent itemset is also considered and two data sanitization approaches have been developed. Based on the experimental results, it can be observed that the item conflict ratio based sanitization approach minimizes non-sensitive itemsets missed and modifications in the original database. To summarize, the research works devised data transformation approaches by which privacy was ensured while maintaining accuracy in data mining