Developing Machine Learning Prediction Models for Construction Material Prices in Nigeria

Abstract

This study aims to develop machine learning prediction models for construction materials susceptible to price fluctuations in Nigeria. Data relating to construction material price influencing factors and construction material prices were obtained from Nigerian Bureau of Statistics (NBS), Central Bank of Nigeria (CBN), and vendors. Making use of Python programming language on Spyder version 3.6 software, a combination of Back Propagation Neural Network (BPNN) and Autoencoder was utilised for data training/model development. The developed models' predictive performance was validated by comparing predicted and actual prices of building material prices using mean-square error (MSE). Results revealed that the developed Autoencoder-BPNN model had an accuracy ranging from 82.04% to 96.92% and was found to be the best for reinforcement. While the BPNN only model, on the other hand, had accuracy ranging from 92.87% to 98.69% and was found to be the best for steel and cement. The models are expected to assist both client quantity surveyors and contractors in coming up with more accurate estimates of construction material prices for efficient cash flow management of construction projects in Nigeria. The developed models put forward a new course for predicting future prices of construction materials most susceptible to price fluctuations in Nigeria.