2
$\begingroup$

My data has timestamps corresponding to the failure occurrences of a specific component in machinery. The timestamps are not uniformly distributed. My question is: 1) what methods can I use to (almost) accurately to forecast future occurrences (timestamps) of Failure? 2)What other features can I derive?

What I've tried so far:

Since the timestamp sequence is unevenly spaced I've derived a feature datediff= difference between sequential fault occurrences. Since it is now a univariate time-series I have tried classical time-series forecasting methods like ARIMA and SARIMA (hasn't worked out well)

I am posting the seasonal decompositions of the time-series freq=7(weekly)

weekly freq=30 monthly acf/pacf acf

$\endgroup$
0
$\begingroup$

I reccomend you to use Prophet:

Prophet is a procedure for forecasting time series data based on an additive model where non-linear trends are fit with yearly, weekly, and daily seasonality, plus holiday effects. It works best with time series that have strong seasonal effects and several seasons of historical data. Prophet is robust to missing data and shifts in the trend, and typically handles outliers well.

In the documentation you can see that is really easy to implement in python.

Also you could convert the problem to a supervised learning one. You can read this blog where they make an introduction of how to face the problem.

| improve this answer | |
$\endgroup$
  • $\begingroup$ Could you tell me how that would solve my problem? $\endgroup$ – Devarshi Goswami Mar 6 at 11:36
  • $\begingroup$ @DevarshiGoswami instead of using ARIMA, prophet should be able to have better results $\endgroup$ – Carlos Mougan Mar 6 at 13:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.