In paper [2], Mollestad and Skowron propose an algorithm excavating the default regular from the inconsistent data. But,the algorithm can't fitter noises effectively and Its operation is large,needs too much time and Its efficiency is lower because its "up and down" searching strategy begins to search unavoidably from the upper layer which includes the most attributes. So,the paper first proposes the method of determining the attribute weight. ON the basis of the method,the paper defines the concept of the weight regular support degree and the concept of the weight regular trust degree and givs the MDWRBR algorithm,which can filter noises effectively and determine the searching direction and the stop condition and can end the operation before the regular mining conducts to the upper layer. So the algorithm reduces the operation and saves time and has some practical value.
