The full utilization of the past data, the stability of the external environment, the stability of the internal correlation variables.

Finally, the main variables determine the reliability of the prediction.

Dear Yin Zhe,

I'm interesting by vibration signal  ( Working with IMS Data  - Run to Failure).

Bearing Data Set in https://ti.arc.nasa.gov/tech/dash/pcoe/prognostic-data-repository/

Thank you

Dear Mohamed Chlalouli,

Here are some references that addresses the feature selection regarding prognostics.

There are many approaches that addresses this topic. You can find a lot of metrics to assess the "prognoticability" of your features. You can calculate metrics such as: Monotonicity , Prognosability , and Trendability defined in this paper: Coble, J., & Hines, J. W. (2009, September). Identifying optimal prognostic parameters from data: a genetic algorithms approach. In Annual conference of the prognostics and health management society, san diego, ca, september. http://ftp.phmsociety.org/sites/phmsociety.org/files/phm_submission/2009/phmc_09_69.pdf

Or separability criteria defined in this paper:

Medjaher, K., Camci, F., & Zerhouni, N. (2012, July). Feature extraction and evaluation for Health Assessment and Failure prognostics. In Proceedings of First European Conference of the Prognostics and Health Management Society, PHM-E'12.

Or go further with feature fusion, in this paper:

Atamuradov, V., & Camci, F. Evaluation of Features with Changing Effectiveness for Prognostics.

Thank you dear Mehdi,

I find your response so useful as well as the references you cited.

I'll take a deep look on that.