Robust Combination for K-Nearest Neighbors’ Forecasting Under Large Fluctuant Short-Term Highway Traffic Flows

To improve the forecasting effect for large fluctuant traffic flows and shorten the computation time in fore-casting short-term highway traffic flows,two improvements were made to the K-nearest neighbors(K-NNs) non-parametric regression method.First,instead of distance or norm,correlation coefficient was used to select K-NNs.Second,the variance reciprocal method in combination forecasts was adopted to synthesize the outcome by K-NNs.The K-NNs selected by correlation coefficients were better than those by distances.Robust variance estimators were employed to suppressing the negative impacts of outliers.The above improvements reduced the search time and in-creased the accuracy by about 1%.The improved effects are confirmed by the numerical test results.