Current Proceedings on Technology
Yazarlar: Elham Parvinnia
Konular:-
Anahtar Kelimeler:Adaptive distance measure,Genetic algorithm,Nearest neighbor,Prototype reduction
Özet: The Nearest Neighbor (NN) classifier uses all training instances in the generalization phase and causes slow execution speed and high storage requirement. In this paper we propose a two-step algorithm that tries to reduce instance sets, decrease the effect of noisy data, and increase classification accuracy. In the first step, the algorithm selects some prototypes by a genetic algorithm. In the next step, the algorithm uses a changed version of an adaptive distance measure to assign weight to selected prototypes. Experimental results show that the proposed method has significant considerable performance in NN classification accuracy, and instance reduction in compare with similar previous methods.