Thingworx analytics - Overpredicted VS Underpredicted

[seanccc]

Hi, 

I'm using thingworx analytics to predict the remaining useful life for a device , and I got the attached result.  For the case of RUL,  I would prefer Underpredicted to Overpredicted ( predictive RUL < actual RUL). 

How can I tuning the super paramters of learner or maybe train data as well to make the model tends to more Underpredicted ? 

Regards,

Sean

[jgreiner · Accepted solution]

Hi Sean,

I have reached out to some of my colleagues with your question to see if they had any advice on how to adjust configurations to create the desired model.  

One tip that I can give you now is to try training several different models and making sure to use a different configuration of learning techniques, Ensemble Techniques, and comparison metrics and seeing if that will create a model that will work for your use case.  These changes can be made under the Advanced Model Configuration tab when creating a new model.

Let me know if you have any further questions or concerns.

Warm Regards,

John