Artificial Neural Network for Cost Prediction of Drainage Infrastructure in a Region with Limited Project Data

¹ Master of Civil Engineering, Faculty of Engineering, Universitas 17 Agustus 1945 Surabaya, Indonesia
² Supervising Lecturer, Master of Civil Engineering, Faculty of Engineering, Universitas 17 Agustus 1945 Surabaya, Indonesia
³ College of Engineering, Nihon University, Japan

^* Corresponding author: rahmatck39@gmail.com

Abstract

Classical estimation methods in infrastructure planning for drainage system in budget constraint area like South Buton Regency of Indonesia are frequently inadequate as they fail to properly simulate complex nonlinear interdependencies between the different influencing cost factors thus leading to significant budget overruns. In this study, an ANN cost estimation model specifically built for drainage works projects is introduced. In order to overcome the difficulty of a limited number of historical data, we employed a purposeful data augmentation technique and staged over an extensive training dataset. The best ANN structure was gradually adjusted and developed, leading to a deep multilayer perceptron, consisting of three hidden layers (32-16-8) with 7 input factors that represented as the architecture for the model. The proposed model performance was rigorously tested using 5-fold cross-validation resulting in an excellent MAPE of 3.04%, R² = 0.998 and low values of error for MAE and RMSE. These numbers show how incredible the model is in predicting results and its trustworthiness. Comparison with Multiple Linear Regression and Random Forest models proved the superiority of the proposed deep ANN architecture. This study is able to recommended that the ANN model developed could effectively be employed for computing reliable cost estimates under low available data conditioned situations, which facilitates proper utilization of public funds.