KARPAGAM Journal of Computer Science (ISSN : 0973-2926)

Performance Modelling Based on Benchmark Data for Microcomputer Hardware
Authors : Ibrahim Akman & Yasar Yesilcay

This paper proposes the use of selected multivariate methods to evaluate performances of the hardware of microcomputers using their performance data, speed and price. The evaluation is done by classifying the PCs into different categories in terms of level of their performances. In order to do the evaluation, the cluster analysis and discriminant analysis methods are used in sequence. The cluster analysis is a technique that uses selected characteristics of n objects to group them into mutually exclusive classes, or clusters, so that "similar" objects are in the same cluster. The discriminant analysis produces a mathematical model, called discriminant function, to determine how a new object is assigned to one of given classes of similar objects. This function is extracted using independent variables corressponding selected characteristics of objects and using the classes obtained from cluster analysis. This study uses hardware benchmark data in the cluster analysis to group the PCs into classes of PCs with similar performances. The discriminant analysis is then developed by using benchmark data, speed and price as independent variables and groups of clusters. Elementary statistical mesasures are also associated to extract some descriptive results as a part of the analyses. The performance of proposed method is demonstrated with data from 173 models of different PC brands. The discriminant function obtained is shown to classify PCs according to their performances with high probability of correct classification, namely 94.8%.