Reverse Principal Component Analysis for Multi-Output Regression

The problem of multi-output regression deals with predicting more than one value given an observation. This paper proposes a novel method to accomplish this task by using a popular technique named Principal Component Analysis (PCA). The approach is to reduce the dimensions of the target data and make predictions on it, following which the predictions are transformed to the higher dimension. This approach was compared against several existing approaches using publicly available datasets. It was found to largely outperform other approaches. Application areas include (not limited to) climatology, genetics, image processing and computer vision, and medicine.