Predicting construction worker tasks from location data
While the workers are completing construction tasks, ultra-wide band technology tracks their location in the construction site. After a task is complete, we harness the power of machine learning to generate a short list of which kind of task it might have been. Task entry is just a simple tap away.
Today's large-scale construction projects require a complex system of labor tasks to finish the job. Accurate tracking of which tasks have been completed is of utmost importance for scheduling and cost-estimation, but complex jobs have hundreds of possibly over-lapping task codes which can be impossible to keep track of.
Our tech automatically suggests the most-likely codes the instant someone finishes a task. That way, you can know exactly what's been done and what's left to do at all times.
An Ultra-Wide Band, Real Time Location System pings each worker's phone every second and records a location. These precisely-timed measurements give the system the power to accurately determine where the workers are and even how fast they are moving.
65% of the time, we suggest the only a single taskcode. More than 95% of the time we suggest three codes or fewer. These suggestsions nearly always contain the right taskcode.
A few minutes of location data tells a lot about what kind of work is getting done.
These results were possible with about ten different tasks, dozens of different workers and dozens of different individual working locations. Some tasks are performed many times in each room, others not at all.
Many machine learning approaches can't handle large amounts of missing data. We use a carefully designed algorithm to ensure that even significant gaps in data recording don't compromise the results of the system. This allows us to increase our training data size and make accurate predictions with real-world data.
