Frequency Ratio: a method for dealing with missing values within nearest neighbour search

In this paper we introduce the Frequency Ratio (FR) method for dealing with missing values within nearest neighbour search. We test the FR method on known medical datasets from the UCI machine learning repository. We compare the accuracy of the FR method with five commonly used methods (three "imputation" and two "bypassing" methods) for dealing with values that are "missing completely at random" (MCAR) for the purpose of classification. We discovered that in most cases, the FR method outperforms the other methods. We conclude that the FR method is a strong addition to the commonly used methods for dealing with missing values within the nearest neighbour method.