Penalized angular regression for personalized predictions

Personalization is becoming an important aspect of many predictive applications. We introduce a penalized regression method which inherently implements personalization. Personalized angle (PAN) constructs coefficients that are specific to the covariate vector for one producing prediction, thus personalizing model itself. This achieved by penalizing normalized prediction given vector. The therefore penalizes coefficients, or angles in hyperspherical parametrization, introducing new angle-based class penalties. PAN hence combines two novel concepts: and For orthogonal design matrix, we show estimator solution low-dimensional eigenvector equation. Based on construct efficient algorithm calculate estimator. propose parametric bootstrap procedure selecting tuning parameter, simulations can outperform ordinary least squares, ridge other methods terms error. Finally, demonstrate medical application.