Implementation of Unsupervised Learning Method in Rule Learning from Construction Schedules

The construction industry has insufficient utilization of standard work and workload. Generally, scheduling for construction projects follows the common sense of the industry. The sequencing of activity and its duration estimation is highly dependent on the experience of the experts who are assigning them to the project, and it is a considerable barrier for automating scheduling process. To overcome this challenge, the FP-Growth algorithm, which is an automated unsupervised learning tool, applied to create a platform for the acquisition of knowledge from actual construction schedules which are the outcome of experienced experts. The main advantage of this method in comparison to supervised learning models is the fact that it can generate contractor-specific rules from a given schedule and also identify a variety of potential path when it is applied for multiple projects which are similar to each other. The main contribution of FPGrowth Algorithm to this research is in finding association rules between sets of activities and identifying recurrent patterns in the sequence of activities, their duration, logical relationship (FF, SS, SF, FS) and specifications in different sections of construction projects. The model applied on schedules of two case studies with different occupational function and structural material. The model substantiated to be capable of learning and identifying various rules including activity durations, predecessor activity and logical relationship and lead times that can happen in between two related activities.