Using the principal component analysis method as a tool in contractor pre‐qualification

Abstract

Contractor pre‐qualification can be regarded as a complicated, two‐group, non‐linear classification problem. It involves a variety of subjective and uncertain information extracted from various parties such as contractors, pre‐qualifiers and project teams. Non‐linearity, uncertainty and subjectivity are the three predominant characteristics of the contractor pre‐qualification process. This makes the process more of an art than a scientific evaluation. In addition to non‐linearity, uncertainty and subjectivity, contractor pre‐qualification is further complicated by the large number of contractor pre‐qualification criteria (CPC) used in current practice and the multicollinearity existing between contractor attributes. An alternative empirical method using principal component analysis (PCA) is proposed for contractor pre‐qualification in this study. The proposed method may alleviate the existing amount of multicollinearity and largely reduce the dimensionality of the pre‐qualification data set. The applicability and potential of PCA for contractor pre‐qualification has been examined by way of two data sets: (1) 73 pre‐qualification cases (37 qualified and 36 disqualified) collected in England and (2) 85 (45 qualified and 40 disqualified) pre‐qualification cases relating to 10 public sector projects in Hong Kong. The PCA‐based results demonstrated that strong and positive inter‐correlations existed between most of the qualifying variables, with the minimum correlation coefficient being 0.121 and the maximum being 0.899, and that qualified and disqualified contractors could be satisfactorily separated.

Keywords:

• Contractor pre‐qualification
• neural networks
• principal component analysis

Acknowledgement

The work described in this paper was fully supported by a grant from the Research Grant Council of the Hong Kong Special Administration Region, China [ Project No. 9040922 (cityU 1293/04E)].