A novel feature engineering-based hybrid approach for precise construction cost estimation using fuzzy-AHP and artificial neural networks

The integration of Artificial Intelligence (AI) into construction cost estimation is transforming the way engineers tackle complex challenges. AI algorithms play a pivotal role in predicting project costs by scrutinizing historical data from previous construction projects. This study introduces a novel hybrid approach combining the Fuzzy-Analytic Hierarchy Process (F-AHP) and Artificial Neural Networks (ANNs) to enhance the accuracy of construction cost estimations. The F-AHP identifies the most influential features affecting project costs and assigns appropriate weights, while the ANN captures the intricate and ‘unknown’ mathematical relationships between these features and the corresponding costs. The approach was validated using real-world data from 96 residential and specialized housing projects in Amman, Jordan spanning from 2014 to 2021. Eleven key features are selected to train the ANN prediction model. To enhance prediction accuracy, the ANN model is optimized in terms of training/learning algorithms, hidden-neuron activation functions and the number of hidden neurons. The Qusai Newton, an exponential linear unit (Elu) function and 8 neurons were identified, respectively, as the optimal configuration, based on standard accuracy metrics. This optimized setup yields a Root Mean Square Error of 31,000 JD, a Mean Absolute Percentage Error of 3% and a maximum Coefficient of Determination of 98%.