Model adjustment for predicting time to failure

[mahsarouz]

Hi,

I just have few questions regarding this problem (time to failure prediction). I have few thoughts and I need your feedback about them (https://github.com/Microsoft/SQL-Server-R-Services-Samples/blob/master/PredictiveMaintenance/R/02a-regression-modeling.R)

Do we have right censoring in this problem? or did you assume that all the observations died and you know the exact time of the failure? In the case of right censoring, I was thinking classic ML models such as boosted tree and decision forest might not be appropriate and we need to focus on survival models. In case of considering only the devices that did die, there might be a bias in the model as we didn't consider those who are still alive. I am just trying to understand the problem better. I appreciate if you could possibly explain a little bit more about the problem definition and the model adjustment.

Thanks,
Mahsa