Optimization of Elman Neural Network Using Genetic Algorithm for Construction Cost Estimation and Overspending Risk Analysis

Abstract

This study proposes a model based on the Elman neural network and improves it using a Genetic Algorithm (GA) to increase the accuracy of construction cost estimation and accurately analyze the overspending risk. First, an index system containing multiple dimensions such as building features, structural features, project positioning, and project environment is constructed to comprehensively capture the key factors affecting construction cost and overspending risk. Second, the Elman neural network’s structure and operation are thoroughly examined, and the GA optimizes the network’s weights and thresholds to improve the model’s predictive power. On the training set, the optimized GA-Elman model demonstrates great prediction accuracy, with relative error (RE) percentages between predicted and true values typically falling within ±1%. On the test set, the GA-Elman model performs better than the original Elman model in both difference and RE, with a Mean Absolute Percentage Error of 2.75%, a decrease of 18.4% compared to the Elman model. These results indicate that the GA-Elman model is more accurate in cost prediction and more effective in identifying potential overspending risks. This study provides a powerful tool for cost control and budget management in the construction industry and a new perspective on the application of neural networks in construction economics.