关键词: 软件可靠性预测, 软件度量, 属性选择, 遗传算法, KNN算法

Abstract: Aiming at the the question of dimension disaster in software reliability prediction，a new method of software metrics selection based on Adaptive Genetic Algorithm（AGA） and K-Nearest Neighbor（KNN） algorithm is put forward.It is used to filter a pivotal metrics subset which is super correlative with software reliability.It uses genetic algorithm to search the metrics subset randomly and uses the classification error rate of KNN classifier as the learning algorithm and evaluation method.The simulation results show the dimension of the pivotal attribute selected by AGA-KNN is obviously smaller than the original metrics set，and the pivotal attribute subset can make the classification accuracy higher than used all the metrics data.

Key words: software reliability prediction, software metrics, attribute selection, genetic algorithm, K-Nearest Neighbor（KNN）