A Study of Neuro-Weighted Nearest-Neighbour Classification

Irrelevant features of the dataset have a certain impact on the judgment ability of the classifier. Effectively distinguishing strong and weakly relevant features can improve classification performance. Since neural networks can effectively dig out the underlying impact of the conditional features on the decision, in this work, the knowledge learned and stored in the parameters of neural networks will be applied to weight the features. Furthermore, the weighted features will be used in the similarity-based nearest-neighbour (SNN) classifier to enhance the model's classification performance. The experimental results demonstrate that the proposed weighted nearest-neighbour approach generally outperforms many popular classification techniques.