An Artificial Intelligence Based Model for Construction Activity Sequence Prediction in Highway Projects: Fid-Bau Portal

An Artificial Intelligence Based Model for Construction Activity Sequence Prediction in Highway Projects

Alikhani, Hamed / Jeong, H. David

Activity sequence determination plays a crucial role in construction scheduling. Many studies have been conducted to leverage sequence knowledge from historical schedules. However, they still lack in fully capturing the sequence knowledge due to the inherent unstructured characteristics of sequential patterns. This research applies an advanced Machine Learning technique called Long Short-Term Memory Recurrent Neural Network on historical schedule data to learn the activity dependency and sequence logic. The study collected 311 as-built highway schedules from a state department of transportation. Word Embedding approach is used for semantic annotation of activities to map them to feature vectors. Features based on work type and resource utilization are assigned to each activity. The model is trained to take a sequence of schedules and predict the most likely successors. The model validation yielded the accuracy of 80%. The activity feature vectors are used to visualize the project schedules. The contributions of this paper are (1) the visualized information of project activities can significantly help schedulers easily identify similar projects, and (2) schedulers can use the successor activity prediction capability of the model in activity sequence determination in developing pre-construction schedules and during construction update schedules.