Benjaoran, V, Dawood, N and Scott, D (2004) Bespoke precast productivity estimation with neural network model. In: Khosrowshahi, F (Ed.), 20th Annual ARCOM Conference, 1-3 September 2004, Heriot Watt University. Association of Researchers in Construction Management, Vol. 2, 1063-74. BESPOKE PRECAST PRODUCTIVITY ESTIMATION WITH NEURAL NETWORK MODEL Vacharapoom Benjaoran 1 , Nashwan Dawood, and Darren Scott Centre for Construction Innovation Research, University of Teesside, Middlesbrough, TS1 3BA, UK Bespoke precast-concrete components are custom made for construction projects. The variety of product designs results in requiring different manufacturing time. To estimate the productivity of four precast routines, this study identifies twenty influential factors based on the difficulty in product designs and manpower. These influential factors are such as nominal height, length, and width, tiling area, the number of curves, the number of embedded parts, concrete strength, slump, reinforcement weight, and the number of different bar shapes, etc. Productivity estimation models are formulated using two techniques: neural network (NN) and multivariable linear regression (MLR). The estimation performance from both techniques is measured with three statistical values, namely absolute percentage error, mean square error, and correlation coefficient. The experimentation results show that MLR gives insignificantly better performance than NN. However, standardised residuals from the NN are distributed in the narrower range than the ones from the MLR. Keywords: bespoke precast-concrete production, multivariable linear regression, neural network, productivity estimation. INTRODUCTION Estimation is a necessary assignment in construction management. It includes cost (bid preparation, budget), time (productivity, project schedule), or quality estimation. Despite from that, the estimation is complicated, intuitive and approximate. For the productivity estimation, there can be so many factors that influence the productivity of construction tasks because the tasks involve long sequential processes, craftsmanship, many materials and tools, and changeable site conditions. Some of the factors are easily recognised; some of them may not. Also, the extent of these factors affect the productivity is difficult to identify. To avoid these problems, an empirical estimation technique has been used. The technique is based on estimators’ experience to consider current task conditions and to adjust a standard productivity figure in a handbook for a suitable value. It is a simple technique but lacks of consistency and learning from the past cases (Chao and Skibniewski, 1994). For bespoke precast-concrete production, precast components are custom made for a construction project. The construction tasks are brought into a factory where there is a controlled working environment. The productivity estimation of precast manufacturing is strongly required for arranging a production schedule. Factors that are identified to affect the productivity are the difficulty in custom product designs. This paper reports a study of the estimation of bespoke precast productivity using two computing techniques: neural network (NN) and multivariable linear regression (MLR). The productivity model with neural network is developed for bespoke precast 1 b.vacharapoom@tees.ac.uk