I often am asked the simple question, “what makes a good APR modeler?” For those who are reading this and don’t know the term APR, it stands for Advanced Pattern Recognition, and is a term that refers to the collection of predictive analytics models that belong to a class of artificial intelligence approaches made up of unsupervervised and supervised learning, and in some cases hybrid approaches. The APR models produce an expected value for each actual variable in the model. Deviations are trended to indicate early warning of anomalous conditions.
First, understand that the APR technique does not require 30 or more years of subject matter expertise on every piece of equipment that is to be modeled. Though helpful and often necessary for model validation, deep specific expertise is less important to a model developer than general ability to think outside the box and apply sound practices to data analysis and model structure.
Second, experience with data analysis – and, in particular - drawing conclusions from data trends where only partial information is available. Understanding sources of process variation is the key to both building good APR models and interpreting results.
Third, the ability to work collaboratively with experts on equipment, data scientists, managers who love numbers, and peers is an essential part of daily life that should not be overlooked.
Comments or questions? Write me at firstname.lastname@example.org.