Predicting Galaxy Ellipticity to Reduce Shape Noise in Weak Lensing Research

We predict the ellipticity of galaxies from a variety of variables available through data from the Canada-France-Hawaii Lensing Survey. A survey of regression methods, including both nonparametric and tree-based methods, are applied to the prediction problem and results are discussed and compared. We present graphical methods for finding interactions between predictors. Leaveone-out cross-validation, K-fold cross-validation, and the Akaike information criterion are used to choose parameters of the models. We find that projection pursuit regression performs the best on a seperate test set of data. Tree-based models performed poorly due to strong linear relationships in the data.