An Algorithm of Mining Exceptions in High Dimensional Data Cube

Existing algorithms of mining exceptions in high dimension data cube often require users to fix out one or more thresholds beforehand to judge whether a data point is an exception, and different thresholds will lead to different efficiency. To overcome this problem, we propose a new definition of exception, interval exception, which can help users find the exceptions in data cell fast and efficiently without any threshold users given. This method computes reference range of a point firstly according to regression parameters, and then considers whether it is an interval exception by comparing its absolute value of residual to reference range. Performance study shows that the method is practical and effective.