Machine Learning-Based Modelling for Predicting Construction Maintenance Costs Using Multiple Linear Regression and Random Forest

Cahya Esther Purnama Wulan Esther (Postgraduate Student at Institute Teknologi Sepuluh Nopember, Surabaya, Indonesia)
Yusroniya Eka Putri Rachman W. (Postgraduate Student at Institute Teknologi Sepuluh Nopember, Surabaya, Indonesia)
Farida Rachmawati (Postgraduate Student at Institute Teknologi Sepuluh Nopember, Surabaya, Indonesia)

ARMADA : Jurnal Penelitian Multidisiplin
Vol. 4 No. 7 (2026): ARMADA : Jurnal Penelitian Multidisplin, July 2026

During the tendering phase, contractors must at least estimate the whole project’s costs. However, contractors often focus primarily on the execution phase, whilst the maintenance and handover phases are not adequately accounted for. Such behaviour risks reducing profits or even resulting in losses. This study aims to develop a data-driven maintenance cost prediction model, specifically for water infrastructure projects, and to compare the performance of the Multiple Linear Regression (MLR) and Random Forest Regressor (RF) prediction models. MLR and RF are machine learning-based methods with contrasting data processing characteristics. MLR processes data linearly, whilst the RF processes data randomly. The RF and MLR models utilise raw data and logarithmic transformations. This study produced  model using data from 18 projects. The best model performance on the 18 data points was achieved by RF 10 – Log – KFold, with the lowest MAPE (21.19%), MAE (5.25×10⁸) and RMSE (9.2×10⁸).  In this study, the prediction model using the RF algorithm performed better than the MLR algorithm. However, the findings of this study cannot be generalised due to data limitations, which appear to have influenced the results.