We present an extension of the usual agent-based data mining cooperative work flow that adds a so-called adjustment work flow. It allows for the use of various knowledge-based strategies that use information gathered from the miners and other agents to adjust the whole system to the particular data set that is mined. Among these strategies, in addition to the basic exchange of hints between the miners, are parameter adjustment of the miners and the use of a clustering miner to select good working data sets. Our experimental evaluation in mining rules for two medical data sets shows that adding a loop with the adjustment work flow substantially improves the efficiency of the system with all the strategies contributing to this improvement.
Generated: 18/8/2013