Prediction Intervals of Principal Component Regression with Applications to Molecular Descriptors Datasets

The prediction interval of the Principal Component Regression (PCR) model is usually based on some strong distributional assumptions. To overcome this drawback, this study extends a prediction interval estimation method by considering a condition of distribution-free. Six different prediction interval estimation methods are developed for constructing prediction intervals for PCR models. The simulated and real data experiment results show that the developed methods perform better than some state-of-the-art methods. This study can enrich the tools of PCR model prediction and statistical inference and is significant for data analysis.