Comparative Study of Cost Significant Model and Artificial Neural Networks Methods for River Retaining Wall Cost Estimation in Grobogan Regency

Abstract

Grobogan Regency in Central Java Province has a high level of flood risk, so the construction of river retaining walls is an important infrastructure for disaster mitigation. The estimation of construction costs at the early planning stage plays a crucial role in budgeting and technical decision-making. This study aims to compare the accuracy and consistency of two cost estimation approaches: Cost Significant Model (CSM), based on multiple linear regression, and Artificial Neural Networks (ANN) using the backpropagation algorithm. The dataset comprises 42 Bill of Quantity (BoQ) documents (37 training data and 5 testing data), with additional validation conducted through field surveys at seven proposed retaining wall locations. Model performance was evaluated using Mean Absolute Percentage Error (MAPE) to measure accuracy and Bland–Altman Plot to assess consistency. The results indicate that CSM achieved a MAPE value of 1.70%, which is lower than that of ANN, which yielded 2.50%. The Bland–Altman analysis also shows that CSM demonstrates higher consistency, as the linear regression approach allows prediction beyond the training data range, making it more adaptive to actual conditions. In contrast, ANN tends to be constrained within the normalized training data range, reducing its flexibility when encountering new data variations. Therefore, it can be concluded that CSM performs better than ANN in terms of accuracy and consistency in estimating the construction cost of river retaining walls in Grobogan Regency.