Predicting 3D Printer Failure

As part of a data science practicum at Formlabs, I leveraged product usage data and a cascading ensemble of decision trees to predict if a printer would be sent back under the company’s Return Merchandise Authorization (RMA) program and, more specifically, under which of the 19 known symptoms it would be classified.

If this model were to be deployed as tested by the holdout dataset, the business would recover at least 39.5% of every dollar lost by an RMA if it were to expedite preventative measures before failures happened.


As a Fellow at Formlabs, arguably the most innovative 3D printing in the country, I leveraged product usage data to not only understand, but rather predict product failures. Implemented, my pipeline would lead to a 40% recovery of costs associated with returned products.