Abstract: Decision is an effective classification method．But during the building process of decision tree,there usually appear over-fitting phenomena of models．This paper discussed soft pruning processing by using BP pruning which is based on BP neural network．Then,according to the shortages of BP pruning,this paper proposed a revised algorithm,named GBP-Pruning．This algorithm is able to train weight and threshold value of BP-Pruning model by bringing in genetic algorithm,so that it can overcome the shortages of BP-Pruning．It also proved the feasibility of GBP-Pruning.

Key words: Data mining,Decision-tree,BP neural network,Genetic algorithm,Pruning algorithm