Balanced-kNN: A New Lazy Learning Algorithm and its Evaluation
Itaru Nagayama, Akira Miyahara, Koichi Shimabukuro · IEEJ Transactions on Industry Applications · 2019
This paper proposes a new lazy learning algorithm, named balanced-kNN, for high performance robust classification of noisy patterns. K-nearest neighbor (k-NN) is a simple and powerful method with a high accuracy for various real world applications using unbiased datasets. However, noisy datasets are often gathered in real world applications. This paper presents a new robust algorithm, balanced-kNN, and compares the prediction accuracy with some conventional methods by using UCI datasets. The experimental results show that the balanced-kNN algorithm can perform more efficient classification of noisy data than the normal-kNN and weighted-kNN algorithms.