Evaluation of modified PLS regression method to fill the missing values in training dataset

In This paper, a new method for handling the missing data in the Partial Least Squares (PLS) regression method is proposed. The idea to handle missing data is by filling the empty field internally by the regression among predictors. First the PLS regression model is built on the data with no missing values which will have a predictor set and a response set. Then we consider the data with missing values in the predictor set and we fill in the missing data by using the internal regression models built with in the predictor set. After filling the missing values we compute the PLS regression model for this new data set. We use same test data set to evaluate these two built models. The evaluation is based on the R-square values and difference between the measured and predicted values.