Comparing advanced and traditional machine learning algorithms for construction duration prediction: a case study of Addis Ababa’s public sector

Abstract

Accurate prediction of construction durations is crucial for effective project management, particularly in rapidly urbanizing areas such as Addis Ababa. However, there exists a notable research gap regarding the comparative analysis of advanced machine learning (ML) algorithms against traditional methods. This study aims to develop and evaluate various advanced ML algorithms to predict construction completion times in Addis Ababa, with the goal of improving resource allocation and enhancing client satisfaction. Data were collected through surveys administered to multiple construction organizations within the city, which served as the foundation for training, validating, and comparing a range of ML models. The research utilized the caret package in R for model development and assessment, incorporating methodologies such as artificial neural networks (NN), Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Classification and Regression Trees (CART). To evaluate variable importance, multivariate visualizations, including correlation and scatter plot matrices, were employed, while performance metrics such as mean absolute error (MAE), root mean square error (RMSE), and R-squared (R²) were utilized for model comparison. The findings indicated that the RF model achieved an RMSE of 74 days and an R² of 0.97, while the KNN model also demonstrated strong performance with an RMSE of 81 days and an R² of 0.97, marking them as the most accurate models for predicting construction durations. In contrast, the NN model exhibited subpar performance, likely due to constraints related to training data and variable selection. As a result, the RF model was further optimized to improve its predictive accuracy. The study concludes that while the RF model proves to be highly effective for predicting construction durations in Addis Ababa, there is a critical need to expand the dataset and incorporate additional variables to enhance the performance of deep learning and other ML algorithms in this field.