Forecasting contractor performance using a neural network and genetic algorithm in a pre-qualification model

Purpose - This paper seeks to introduce an evolved hybrid genetic algorithm and neural network (GNN) model. The model is developed to predict contractor performance given the current attributes in a process to pre-qualify the most appropriate contractor. The predicted performance is used to pre-qualify the contractors. Design/methodology/approach - Hypothetical and real-life case studies from projects executed in the Gaza Strip and West Bank were collected through structured questionnaires. The evaluation of the contractor's attributes and the corresponding actual performance of the contractor in terms of time, cost, and quality overrun (OR) were collected. The weighted contractor's attributes were used as inputs to the GNN model. The corresponding time, cost, and quality ORs for the same cases were fed as outputs to the GNN model in a supervised learning back propagation neural network (NN). (The adopted training and testing process to develop a trained model is presented.) The training process, including choosing the topology of the required NN using genetic algorithms, is explained. Findings - The results revealed that there is a satisfactory relationship between the contractor attributes and the corresponding performance in terms of contractor's deviation from the client objectives. The accuracy of the model in terms of mean absolute percentage error (MAPE), R2, average absolute error and mean square error revealed that the model has sufficient accuracy for implementation. The average MAPE for time, cost and quality OR is 15 per cent. Consequently, the model accuracy is 85 per cent. Originality/value - The GNN model is able to predict future contractor performance for given attributes.

目的 - 本文旨在介绍一种改进的混合遗传算法和神经网络（GNN）模型。该模型的开发是为了在对最合适的承包商进行资格预审的过程中，根据当前的属性来预测承包商的绩效。预测的绩效用于对承包商进行资格预审。 设计/方法/途径 - 通过结构化问卷收集了在加沙地带和西岸执行的项目的假设案例和实际案例研究。收集了对承包商属性的评估以及承包商在时间、成本和质量超支（OR）方面相应的实际绩效。将加权后的承包商属性用作GNN模型的输入。在监督学习反向传播神经网络（NN）中，将相同案例对应的时间、成本和质量超支作为输出输入到GNN模型中。（介绍了开发训练模型所采用的训练和测试过程。）解释了训练过程，包括使用遗传算法选择所需神经网络的拓扑结构。 发现 - 结果表明，就承包商偏离客户目标而言，承包商属性与相应绩效之间存在令人满意的关系。该模型在平均绝对百分比误差（MAPE）、R²、平均绝对误差和均方误差方面的准确性表明，该模型具有足够的实施精度。时间、成本和质量超支的平均MAPE为15%，因此，模型准确率为85%。 创新性/价值 - GNN模型能够根据给定的属性预测承包商未来的绩效。