Cross-Validation of Short-Term Productivity Forecasting Methodologies

Abstract

Frequent and regular prediction of productivity is needed to effectively manage construction operations in progress. This need is proven by significant deviations in productivity between estimated values and actual values, and the dynamic and stochastic changes in productivity over time. Five statistical methodologies that are appropriate for contemporaneous time series were cross validated in the present study. Validation was conducted by comparing the performance of forecasting models constructed using the methodologies. Performance was measured by evaluating the residual sum of squares and the correlation coefficients. As a result, univariate time series analysis was found to be the best-performing methodology. The univariate time series model is particularly beneficial in two respects. Using a single series of contemporaneous productivity data in numeric form, it reduces the effort expended in collecting and analyzing data, and improves the objectivity of analysis.