Predicting Construction Costs with BIM and Neural Networks

By Nidhi DhullReviewed by Susha Cheriyedath, M.Sc.Oct 9 2024

A recent article published in Heliyon proposed an intelligent construction cost prediction model based on building information modeling (BIM) and Elman neural networks (ENNs). This BIM-ENN model aimed to improve real-time prediction accuracy by addressing complex factors like supply chain fluctuations, design changes, and labor cost variations. 

Background

Cost management has become crucial in the construction industry due to the rapid development of the global economy and increasing competition. Scientific cost management methods are essential to successfully implementing projects, controlling budgets, and managing profit.

However, inherent complexity and uncertainty in construction projects make traditional methods inefficient and unreliable. These methods often rely on historical data and empirical judgment, which are unable to promptly reflect recent market changes and project progress.

Deep learning has demonstrated outstanding predictive performance in various fields. However, its use for intelligent building cost prediction is still in its infancy. Moreover, the few utilized artificial neural network (ANN) models for cost prediction generally struggle to capture real-time data adequately, resulting in low prediction accuracy.

Alternatively, integrating data visualization in BIM technology with ANNs can predict construction costs more accurately, enabling efficient decision-making and project control in construction. Thus, a BIM-ENN model was developed and validated in this study.

Methods

After data visualization and normalization, the proposed ENN model was optimized using the particle swarm optimization (PSO) algorithm. Subsequently, the data from the BIM model was input into ENN, and the ANN parameters were optimized to predict building construction costs. 

BIM was employed initially to digitize and visualize different building aspects, such as geometry (size and shape), attributes (material strength, density, and cost), components (column and beam connections), and spatial topology (layout). This data was preprocessed for cleaning, removing abnormal values, filling in missing values, etc., to ensure consistency and quality. ENN learned the relationships between construction costs and architectural elements from this data.

The performance of the developed BIM-ENN construction cost prediction model was evaluated using research data collected using the web crawling method. This data was sourced between April 2019 and October 2022 from academic papers, government and industry reports, social media, news websites, publicly available corporate data, patents, and open data platforms. It focused on the artificial costs of hot-rolled third-level seismic rebars in Xi'an City.

The BIM-ENN model’s performance was compared to ENN, backpropagation (BP) neural network, long short-term memory network (LSTM) algorithm, and an algorithm reported by other researchers (Li et al. 2023). The evaluation metrics included fitting effect, root mean squared error (RMSE), prediction accuracy, coefficient of determination (R²), and area under the curve (AUC).

Results and Discussion

The developed model demonstrated superior performance over other reported algorithms in crucial metrics such as RMSE, R², prediction accuracy, and AUC. These results exhibited the model’s efficiency, reliability, practical significance, and application potential in construction cost management.

While a consistent trend was not observed across all performance metrics, the BIM-ENN model achieved the minimum RMSE, consistently below 75, and the R², over 0.95, implying that over 95% of the cost predictions were explainable. Thus, the developed model demonstrated high (requisite prediction standards) efficacy and feasibility in predicting steel bar prices.

Alternatively, the model by Li et al. depicted RMSE values over 95 and R² consistently under 0.95. Additionally, the BP algorithm exhibited the maximum RMSE, over 130, and the lowest R², consistently under 79.51. Thus, the employed PSO algorithm effectively enhanced the BIM-ENN model, mitigating the tendency of a single ENN to converge to a local extreme value.

Considering the AUC values from different models, the BIM-ENN model demonstrated superior specificity and sensitivity across most thresholds. Notably, its sensitivity was higher at higher thresholds, achieving an AUC value of 0.95 and exhibiting strong discriminative ability. Notably, Li et al.’s model maintained good specificity and sensitivity at various thresholds but was lower than the BIM-ENN model.

Alternatively, ENN and LSTM had greater sensitivity at lower thresholds, but it declined with increasing thresholds, while specificity remained relatively high, indicating the possibility of more false positives. The BP algorithm performed least favorably with lower specificity and sensitivity at all thresholds.

Conclusion

Overall, the researchers successfully developed an intelligent building construction cost prediction model combining the BIM technology with ENN. Introducing the PSO algorithm enhanced the prediction model’s accuracy and generalization capability. The developed model achieved an average prediction accuracy of 95.83% across different months, with an RMSE under 75 and an R² over 0.95. These metrics outperformed other reported models.

Thus, the BIM-ENN model can prove to be a precise and reliable cost management tool for the construction industry. The researchers suggest further expanding the applicability of the model to other aspects of construction management, such as quality control, scheduling, and environmental impact assessment.

Journal Reference

Zhang, Y. & Mo, H. (2024). Intelligent Building Construction Cost Optimization and Prediction by Integrating BIM and Elman Neural Network. Heliyon, 10(18), e37525. DOI: 10.1016/j.heliyon.2024.e37525, https://www.sciencedirect.com/science/article/pii/S2405844024135566

Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.