Cross-layer Analysis of Prediction Accuracy in Ad-hoc networks based on Multivariate Logistic Regression

Abstract

With the advent of ubiquitous networking technologies, there has been plentitude of research proposals in the context of cross-layer design for wireless communication systems. This network design approach is particularly important for the implementation of ad-hoc wireless networks with limited channel accessibility, power-constrained operation and variable capacity links. In this paper, we develop a cross-layer design model for the decentralized ad-hoc networks encompassing four layers of the traditional protocol stack architecture based on convex optimization technique. The optimal network attributes thus computed are deployed in the dataset formulation that is fed to the multivariate logistic regression framework. This nonlinear learning methodology is effectively exploited to retrieve the optimal cost function and the optimal set of sigmoid regression parameters associated with each attribute. Finally, to demonstrate its superiority, the prediction accuracy performance of the presented optimization model executed with logistic classifier algorithm is compared with the traditional Levenberg-Marquardt method.