I have a time-series data of air-pressure inside a room. The reading are the output of an physics experiment. The Predictor variable is binary flag which is coded as follows:
If (ending-reading = 0 then 1 else 0)
I have attached the snapshot of the data below. My objective is to predict the likelihood of the ending-reading being 0 for a future time period.
I understand that I can use time-series forecasting like ARIMA or ARIMAX to project the end-reading and then simply refresh the Predictor flag. But I am looking for other alternatives, either supervised or unsupervised methods.
I thought survival-analysis might work but I am not sure if it is applicable in this case since the end-reading can be 0 on multiple days. The experiment doesn't stop if the end-reading on a particular day is 0.
Would logistic regression work on time-series data?
Any help would be much appreciated.
Best Answer
You have asked 2 questions. First if you can use Logistic regression. Of course you can. You can surely have lag values of ending reading/ beginning reading as independent variables. Even ARMA is a kind of regression only.-
http://machinelearningstories.blogspot.in/2016/08/time-series-and-fitting-regression-on.html
Second, As you have state based output values- 0 and 1. You can think of state models for time series like Multi State Modelling (MSM) or Hidden Markov Models ( if data has markov property). In these state models you can have beginning reading, inflow, outflow as external variables and then you can build your model.