Project Dispute Resolution Satisfaction Classification through Neural Network

Abstract

This paper presents an artificial neural network technique of analysis in determining the important factors affecting the outcome of construction dispute resolution processes in Hong Kong. Projects were classified as Favorable or Adverse in terms of Dispute Resolution Satisfaction in accordance with conventional professional practice for deciding on which disputes get resolved. The necessary historical project data sets were collected through structured interview and questionnaire surveys to provide the training details for the building of a Multi-Layer Perceptron artificial neural network. The preliminary analyses conducted indicated that resolution outcome depends on a combination of factors, namely, environment-, organization-, project-, and process-specific. The refinements to the network were achieved through reduction of the numbers of variables and processing elements. Verification of the “Best” network was achieved through the running of a batch function for stabilization. The optimal network so produced was applied to unseen data and achieved a 100% correct testing result for adverse DRS projects. The optimal network also identified design changes as the most critical factor, indicating that projects with a high degree of design changes were more likely to result in dispute requiring the service of alternative dispute resolution techniques or formalized proceedings.