Adaptive linear prediction with application to missing observations

A new method is presented for estimation of missing values for the problem of periodically missing observations. The proposed method is a modification of the conventional linear prediction method in that it uses adaptive order for prediction, it combines both forward prediction and backward prediction, and it allows estimation of missing values to be carried out for multi-passes. It is shown that the new method performs better than the conventional linear prediction method when the number of remaining samples in one period is smaller than the order of the underlying model for the observed signal. An example using real data is illustrated