Application of Artificial Neural Network Methodology for the Prediction of Labor Productivity in the Construction Sector

Abstract

The present research determined the level of prediction of labor productivity in road projects using the methodology of artificial neural networks, due to the relevance of this variable in the construction industry. In order to achieve this objective, first a systematic review was carried out to determine the most influential factors in labor productivity. Then, technical files were compiled that included the following items: "Manual cutting at subgrade level", "Subgrade leveling and compaction with light or manual equipment", "Conformation of granular base" and "Concrete in sidewalks" for the creation of the database used in the training and testing of the neural networks. Then, the optimal model for the development of the final neural networks was evaluated, generating 4 models of Machine Learning Artificial Neural Networks, one for each item, with the training algorithm "Bayesian Regularization" and with 20 neurons in the hidden layer, achieving values of 96%, 97%, 97% and 99% for the correlation factors between input and output values. Finally, the 4 models developed were validated with the application of data from works in progress, demonstrating that the models generated are more accurate and reliable than conventional methods when predicting the real productivity of a batch.