KNN algorithm improving based on cloud model
Yu Liu, Chen Gui-Sheng · 2010
KNN algorithm is particularly sensitive to outliers and noise contained in the training data set. In this paper, we use the reverse cloud algorithm to map the training samples into clouds. Each attribute is mapped to a cloud vector. Reverse cloud algorithm is not sensitive to the noise on data sets and it can eliminate the impact of noise on classification effectively. By comparing the similarity of clouds in the cloud vector, we can calculate the attributes weights. For those attributes with a low weight of properties, we find out merger them to a new attribute which can generate more significant attribute weight than original ones. We present a new KNN algorithm based on Cloud Model and compare our algorithm with classic KNN algorithms and other well-known improved KNN algorithms using 10 data sets. Experiments show that our approach could achieve a better or at least a comparable classification accuracy with other algorithms.