Enhanced Artificial Neural Network Model with Feature Importance Analysis for Drainage Infrastructure Cost Prediction in Data-Scarce Regions

Abstract

Accurate cost estimation of drainage infrastructure is a critical challenge in rapidly urbanizing regions, particularly in areas with limited historical project data. In Indonesia, inadequate estimation methods often lead to cost overruns and inefficient budget allocation, highlighting the need for more reliable predictive approaches. This study aims to develop and validate an Artificial Neural Network (ANN) for predicting drainage construction costs in data-scarce environments. A quantitative research design was employed using data from 10 drainage projects in South Buton Regency, which were expanded to 150 samples through rule-based data augmentation. The ANN model, based on a Multilayer Perceptron (MLP) architecture, was trained and validated using 5-fold cross-validation. Its performance was evaluated using R², MAE, RMSE, and MAPE, and compared with Multiple Linear Regression (MLR), Random Forest (RF), and Support Vector Regression (SVR) models. The results demonstrate that the proposed ANN model achieves superior predictive performance, with an R² of 0.9978 and MAPE of 3.04%, significantly outperforming the benchmark models. Feature importance analysis reveals that material-related costs, particularly stone masonry, are the most influential factors in determining total project cost. The model also shows strong generalizability and robustness across datasets. The findings imply that the integration of ANN, data augmentation, and feature importance analysis provides a practical and scalable solution for cost estimation in resource-constrained regions. This research contributes to improving decision-making in infrastructure planning, enhancing budget accuracy, and supporting more efficient and sustainable public investment strategies.