A Prototype Selection Algorithm Based on Extended Near Neighbor and Affinity Change
Juan Li, Xiaofang Guo · 2016
The condensed nearest neighbor algorithm(CNN) is susceptible to pattern read sequence, abnormal patterns and so on. To deal with the above problems, through the analysis of the relationship between the whole dataset and the individual patterns, a new prototype selection algorithm is proposed based on the extended near neighbor relationship and the affinity changes. First, the proposed algorithm can obtain the detail location information according to the extended near neighbors and affinity value of each pattern. Second, by making full use of these information, the proposed algorithm adjusts the prototype selection strategy. Finally, the prototype updating strategies are adopted to achieve dynamic periodic update of the prototype set. Experimental results show that the final prototype set obtained by the proposed algorithm can better reflect the distribution of the original dataset. Moreover, the proposed algorithm can improve the average reduction ratio while maintaining the better classification accuracy and faster running time than those compared algorithms.