Problem zmiennych zakłócających w agregowanych klasyfikatorach kNN
/ A problem of the noisy variables in the aggregated kNN classifiers

Summary/Abstract: Ensemble learning in discrimination and regression has gained a great appreciation due to the improved stability of the model, and often improved accuracy of the predictions. Aggregating of k nearest neighbors classifiers (kNN), however, faces serious problems. The kNN method, which uses only the distances between objects, is relatively stable, so the diversity of base classifiers can only be achieved by choosing different subspaces. Here, in turn, we encounter the problem of noisy variables, those that do not affect the response variable, and which result in decreasing the accuracy of the kNN classifier. This article reviews the methods of the aggregated kNN classifiers, which were proposed in the literature. We also verify our own proposition of the algorithm. The real data with noisy variables added are used in the empirical study.

• Details
• Contents