Nowadays, massive amounts of data are collected daily in ubiquitous sensor network environments.
With such data available and elaborately structured, it is more important than ever to locate and access knowledge and trends from it using data mining techniques. Those data are valuable to support analyses and decision-making in businesses, for example. Such data normally exist in databases of various types––called ubiquitous databases hereinafter––that might usually be distributed and placed anywhere. A salient problem, however, is that person who engages in data mining using ubiquitous databases would have to spend much time for database selection and data collection, for example, which would be merely a preparatory step to the actual data mining tasks. What a person really
should want must be instead to concentrate on the work of analysis and rule extraction.
In our study, the objective is therefore to develop a virtualization technique so that the data analyst or other user can use all ubiquitous databases as if they were recognized as a single database, thereby helping to reduce the user’s workload.