The K Nearest Neighbors platform predicts a response value based on the responses of the k nearest neighbors. The k nearest neighbors to a given observation are determined by identifying the k smallest Euclidean distances between the predictor values for that observation and the predictor values for each of the other observations. The K Nearest Neighbors platform models both continuous and categorical responses.

A potential drawback of the k nearest neighbors method is that for large scale problems, the prediction formula is often complex and hard to interpret, limiting its usefulness. In addition, K Nearest Neighbors does not calculate probabilities for categorical responses. For more information about the k nearest neighbors method, see Hastie et al. (2009), Hand et al. (2001), and Shmueli et al. (2017).

For a continuous response, the predicted value is the average of the responses for the k nearest neighbors. Each continuous predictor is scaled by its standard deviation. With this scaling, a single predictor with a large range does not excessively influence the distance calculation. Missing values for a continuous predictor are replaced by the mean of that predictor. See Example of K Nearest Neighbors with Continuous Response.

For a categorical response, the predicted value is the most frequent response level for the k nearest neighbors. If two or more levels are tied as the most frequent levels, the predicted response is assigned by selecting one of these levels at random.