Generalized partial functional linear regression for network-linked data

Estimation and prediction methods typically assume that data consist of independent samples. However, in many modern scientific domains, observations often arise from individuals connected through an underlying network. In this paper, we study a generalized partial functional linear regression model for network-linked data and propose an associated estimation procedure. To improve both estimation accuracy and predictive performance, we formulate the problem as the minimization of a penalized loss function, in which network cohesion is promoted via a cohesion penalty. The infinite-dimensional functional predictors and the slope function are approximated using a B-spline basis, and a smoothness penalty on the slope function is also included in the objective function. Simulation studies and a real-data application demonstrate the effectiveness of the proposed method.