A n incremental updating technique is developed for maintenance of the association rules discovered b y database mining. There have been many studies on eficient discovery of association rules in large databases. However, it is nontrivial t o maintain such discovered rules in large databases because a database may allow frequent or occasional updates and such updates may not only invalidate some existing strong association rules but also turn some weak rules into strong ones. In this study, an incremental updating technique is proposed for eficient maintenance of discovered association rules when new transaction data are added to a transaction database. *The research of the first three authors were supported in part by RGC (the Hong Kong Research Grants Council) grant 338 f O65/OO26. The research of the second author was also supported in part by NSERC (the Natural Sciences and Engineering Research Council of Canada) research grant OGP0037230 and an NCE (the Networks of Centres of Excellence of Canada) research grant, IRIS-HMI5. for all the data. Therefore, the number of rules returned from a mining activity could be large. 3. The rules discovered from a database only reflect the current state of the database. To make the rules discovered stable and reliable, a large volume of data should be collected over a substantial period of time.

No abstract

No abstract

Traditional machine learning algorithms assume that data are exact or precise. However, this assumption may not hold in some situations because of data uncertainty arising from measurement errors, data staleness, and repeated measurements, etc. With uncertainty, the value of each data item is represented by a probability distribution function (pdf). In this paper, we propose a novel naive Bayes classification algorithm for uncertain data with a pdf. Our key solution is to extend the class conditional probability estimation in the Bayes model to handle pdf's. Extensive experiments on UCI datasets show that the accuracy of naive Bayes model can be improved by taking into account the uncertainty information.

Most previous research on privacy-preserving data publishing, based on the k-anonymity model, has followed the simplistic approach of homogeneously giving the same generalized value in all quasi-identifiers within a partition. We observe that the anonymization error can be reduced if we follow a non-homogeneous generalization approach for groups of size larger than k. Such an approach would allow tuples within a partition to take different generalized quasi-identifier values. Anonymization following this model is not trivial, as its direct application can easily violate kanonymity. In addition, non-homogeneous generalization allows for additional types of attack, which should be considered in the process. We provide a methodology for verifying whether a nonhomogeneous generalization violates k-anonymity. Then, we propose a technique that generates a non-homogeneous generalization for a partition and show that its result satisfies k-anonymity, however by straightforwardly applying it, privacy can be compromised if the attacker knows the anonymization algorithm. Based on this, we propose a randomization method that prevents this type of attack and show that k-anonymity is not compromised by it. Nonhomogeneous generalization can be used on top of any existing partitioning approach to improve its utility. In addition, we show that a new partitioning technique tailored for non-homogeneous generalization can further improve quality. A thorough experimental evaluation demonstrates that our methodology greatly improves the utility of anonymized data in practice.