Developing an Approach for Multidimensional Data Mining on various Granularities ~ on Example of Financial Portfolio Discovery

Data Mining is considered as a power tool to discover knowledge such as in form of association rule that is useful in business domains for medical diagnosis, Customer Relationship Management (CRM) etc. Yet, there are two drawbacks in conventional mining. Since most of the techniques performs plane-mining based on priori-defined patterns in the data-warehouse as a whole, so a fully re-scan must be done whenever new attributes are added, i.e. multi-dimensional expansion. On the other hand, an association rule may be true on a certain granularity but fail on a larger one. This paper formulates the mining process as a combination of searching for all patterns and matching patterns with a usergiven validity of rules to find. In this paper, an approach based on Concept Taxonomy to produce multidimensional patterns with different granularities in each dimension automatically, and to discover associations for every multidimensional pattern by limited scans than the conventional mining methods. Thereafter, the algorithm is able to deal with multi-dimensional association rules as well as different granularities while given concept-taxonomy and target validity for the rules. Finally, this paper presents experimental results regarding with efficiency and scalability of the algorithm on financial services and data.