Graph Enhanced High Dimensional Kernel Regression

TL;DR

This paper introduces a kernel regression method enhanced with network data to better model nonlinearities and improve predictive accuracy in high-dimensional settings, demonstrated through simulations and real data.

Contribution

It presents a novel kernelized regression model that incorporates network information, extending previous linear models to capture complex nonlinear relationships.

Findings

Significantly improved predictive performance over traditional models

Effective in high-dimensional and network-structured data

Validated on both simulated and real-world datasets

Abstract

In this paper, the flexibility, versatility and predictive power of kernel regression are combined with now lavishly available network data to create regression models with even greater predictive performances. Building from previous work featuring generalized linear models built in the presence of network cohesion data, we construct a kernelized extension that captures subtler nonlinearities in extremely high dimensional spaces and also produces far better predictive performances. Applications of seamless yet substantial adaptation to simulated and real-life data demonstrate the appeal and strength of our work.