How do you predict missing y data values when the level of the y values fluctuates up and down within the data’s range? With ordinary least squares, you must add a term for each change in the slope of y. Hocaoglu, Oysal, and Kurban present a technique to solve this problem: the adaptive neuro-fuzzy inference system (ANFIS). ANFIS combines the learning (adaptive) quality of neural networks with the inference capability of fuzzy systems, to produce accurate predicted values.

This paper discusses how to predict missing average daily wind speeds over a one year time period. The missing wind speeds occur due to downtime for system maintenance or unanticipated system outages.

I must say that I am impressed with the quality of the predictive values generated by ANFIS, but I have a couple of questions. In producing a prediction, ANFIS considers the previous level and the level immediately after the missing data value. Can ANFIS handle missing levels at the endpoints of such a scatter? Also, on page 210, the authors state that for missing data on the fifth day, the wind speeds of the third and fourth days serve as inputs (I think it should be the fourth and sixth days).

This paper is appropriate for anyone who has data with similar characteristics where some of the values are missing.