Mapping Relational Efficiency in Neuro-Fuzzy Hybrid Cost Models

Significant improvements are achievable in the accuracy of cost estimates if cost models adequately incorporate issues of flexibility and uncertainty. This study evaluates the relational efficiencies of the fuzzy composition operators—the max-min and max-product—in establishing the final cost of water infrastructure projects. Cost and project data were collected on 1600 water infrastructure projects completed in the United Kingdom between 2000 and 2011. Neural network was first used to develop relative weightings of relevant cost predictors. These were then standardized into fuzzy sets to establish a consistent effect of each variable on the overall target cost. The strength and degree of relationship of the normalized cost predictor weightings and the fuzzified project attributes were combined using the max-min and max-product composition operators to obtain project cost predictions. The predictions from the two composition operators were compared with the actual cost figures. Results show comparable performance in the efficiency of the composition operators. Based on statistical correlations, the max-product composition operator achieved on average a deviation of 1.71% whereas the max-min composition had an average deviation of 1.86%. Improvements in the relational efficiency of neuro-fuzzy hybrid cost models could assist in developing a robust framework for realistic cost targets on construction projects.