With the large deployment of WiFi networks, indoor localization using WiFi fingerprinting with received signal strength has been widely studied. One of the common localization methods is weighted K-nearest neighbor method (WKNN), which localizes the user to the weighted center of the K best matching points. The performance of this method is affected by the choice of parameter K. Once tuned, this parameter is usually applied to all test samples. In this paper, we study how far localization performance can be improved if this parameter is adapted for different test points. We show with two public access datasets that adapting parameter K for different test points can potentially improve localization performance by over 45% compared to the baseline of only choosing the closest neighbor. Additionally, we analyze the dataset to obtain some stochastic thresholds for dataset filtering and K selection.