This document reports on a research project to develop a set of methods for classifying vendors based on estimated central processing unit performance and to predict that performance based on hardware components.
The authors propose a set of methods to do two things: to classify hardware manufacturers and vendors based on estimated central processing unit (CPU) performance; and to predict computer performance based solely on estimated CPU performance, using a multi-layered neural network in comparison to regression techniques. The authors sought to classify and predict the performance of CPUs based on a set of 10 parameters from an opensource dataset which contained 209 entries representing a variety of vendors and CPU model. The proposed classification method for hardware vendors was the use of classification zones, each of which was labeled with a range of relative performance. The authors’ proposed prediction method was to use multi-layered neural networks, which accounted for nonlinearity in performance data; they also analyzed several neural network architectures in comparison to linear, quadratic, and cubic regression. Results showed that neural networks can be used to obtain low prediction error and high correlation between predicted and published performance values, while cubic regression can produce higher correlation than neural networks when more data is used for training than testing. The authors suggest that their proposed methods can be used to identify suitable hardware replacements.