Duration estimation of a heavy industrial scaffolding project: A case study

Accurate project duration estimation is crucial for effective scheduling, budgeting, resource allocation, and overall construction management. Leveraging historical data from completed projects is an effective strategy to achieve this. In heavy industrial projects, where scaffolding activities can span from thousands to millions of hours, refining the estimation of scaffolding time is vital during the planning phase. This study undertook the analysis of data from a completed heavy industrial scaffolding project, aiming to propose a methodology and models for predicting future projects durations. The proposed methodology not only aids in improved duration but also contributes to cost estimation, scheduling, and project delivery of similar future endeavors. Commencing with data cleaning and categorizing the data based on activity types, the scatter plots of person-hours versus task weight within each category revealed a linear relationship. Consequently, linear models for each category were developed. Statistical factors such as data size, coefficient of determination, and mean absolute error were then utilized to calculate a score for each model, guiding the model selection process which substituted low score models with a parent category with a higher score. The data analysis and modeling were performed five times to ensure robustness and consistency in the results. On average, the initial models yielded a project duration estimate of only 0.36% higher than the actual duration, while the selected models increased this deviation to 4.14%. The scoring and selection process enhances estimation accuracy while maintaining proximity to actual project durations. This research makes three significant contributions: (1) introducing a categorical linear regression approach for scaffolding activity duration prediction, (2) presenting a novel normalization and scoring method that scores models based on statistical factors, and (3) implementing a practical model selection process to substitute weaker models with stronger ones, ultimately strengthening the reliability of activity and project duration predictions.