Development of an Electrical Work Unit Price Analysis System with Random Forest Regressor Approaches

Abstract

The preparation of Cost Budget Plans (RAB) for electrical installation projects is still
commonly performed manually using Microsoft Excel, making cable requirement
calculations and Unit Price Analysis (AHS) time-consuming and prone to errors. This study
develops a machine learning-based system to predict the cable length required for lighting
installations using room characteristics, including the number of rooms, floor area, room
height, and the number of spotlight, downlight, and pendant lighting points. Two
prediction models, Random Forest Regressor and Artificial Neural Network (ANN), were
trained using historical residential electrical installation project data with an 80:20 training
testing split. Model performance was evaluated using R², MAE, MSE, RMSE, and MAPE.
The Random Forest Regressor achieved superior performance with an R² of 0.982 and a
MAPE of 18.94%, outperforming the ANN (R² = 0.942; MAPE = 19.34%). The best model
was integrated into a Streamlit-based web application to support faster, more accurate, and
consistent cable estimation for efficient AHS and RAB preparation.