Neural network model for contractors' pre-qualification for local authority projects

The way in which clients or their consultants undertake to select firms to tender for a given project is a highly complex process and can be very problematic. This is also true for public authorities as, for them, compulsory competitive tendering' is a relatively new concept. Despite its importance, contractors' prequalification is often based on heuristic techniques combining experience, judgement and intuition of the decision makers. This, primarily, stems from the fact that prequalification is not an exact science. For any project, the right choice of the contractor is one of the most important decisions that the client has to make. Therefore, it is envisaged that the development of an effective decision-support model for contractor prequalification can yield significant benefits to the client. By implication, such a model can also be of considerable use to contractors: a model of this nature is an effective marketing tool for contractors to enhance their chances of success to obtain new work. To this end, this work offers a decision-support model that predicts whether or not a contractor should be selected for tendering projects. The focus is on local authorities because, in the absence of a viable universal selection system, there are significant variations in the way they conduct prequalification. The model is based on the use of artificial neural networks (ANN) and uses data relating to 42 local authorities (clients). With the aid of a questionnaire and a scaling system, the prequalification attributes that are considered to be important by clients are identified. The survey indicates significant variations in the level of importance given to different attributes. Statistical methods are adopted to generate additional data representing disqualified instances. Following a preprocessing exercise, the data form the basis of the input and output layers for training the neural-net model. An independent set of data is subjected to a similar preprocessing for testing the model. Tests reveal that the model has a highly satisfactory predictive accuracy and that the ANN technique is a viable tool for the prediction of success or failure of the contractor to qualify to tender for local authority projects.