Rationalizing the Parameters of K-Nearest Neighbor Classification Algorithm

摘要

With arrival of big-data era, data mining algorithm becomes more and more important. K nearest neighbor algorithm is a representative algorithm for data classification; it is a simple classification method which is widely used in many fields. But some unreasonable parameters of KNN limit its scope of application, such as sample feature values must be numeric types; Some unreasonable parameters limit its classification efficiency, such as the number of training samples is too much, too high feature dimension; Some unreasonable parameters limit the effect of classification, such as the selection of K value is not reasonable, such as distance calculating method is not reasonable, Class voting method is not reasonable. This paper proposed some methods to rationalize the unreasonable parameters above, such as feature value quantification, Dimension reduction, weighted distance and weighted voting function. This paper uses experimental results based on benchmark data to show the effect.