1

Problem

I am a software developer and I have been studying ARMA methods. Currently, I have a spatiotemporal stream of data and would like to apply ARMA model to it. I have a binary target (independent variable) comprised of 0 and 1. My problem is that I would like to use ARMA to both detect (classify/estimate) 0 and 1 and then later predict using my stream of data.

Question

What are the ARMA models that can detect/predict/classify/estimate for boolean output (0 or 1)? Is there a specialized name for such a model given my use case is a binary variable as a target? Are there libraries in python or R that I can further study to help me or a literature?

Attempts

I have been unsuccessful in answering this question and have done the following:

  • Introduction to Time Series by Chatfield
  • Maching Learning by Kevin Murphy
  • StackExchange and Google

I beg your pardon and assistance given I am not a domain expert in this area. Any clues, comments, suggested texts, software packages, and materials would be extremely valuable in assisting me further.

Tim
  • 138,066
GeekyOmega
  • 145
  • 1
    Check http://stats.stackexchange.com/questions/197084/binary-time-series or http://stats.stackexchange.com/questions/15068/forecasting-binary-time-series – Tim Sep 20 '16 at 18:30
  • Great links. Thanks! What is the preferred way to measure ARMA models in statistics for classification/detection performance, fit, error etc. In simple terms, what are the metrics to measure model performance and pick a winner? – GeekyOmega Sep 20 '16 at 18:58
  • 1
    There is a whole field of discrete time series models (good reference is: https://www.crcpress.com/Handbook-of-Discrete-Valued-Time-Series/Davis-Holan-Lund-Ravishanker/p/book/9781466577732 another one is http://www.wiley.com/WileyCDA/WileyTitle/productCd-0471363553.html) – bdeonovic Sep 20 '16 at 19:32
  • 1
    might also read about the NDARMA(p, q) models (articles by Weiss) – bdeonovic Sep 20 '16 at 19:33
  • Thanks the responses. I wish I could have given you all more +rep. :) – GeekyOmega Sep 27 '16 at 13:04
  • https://towardsdatascience.com/arima-for-classification-with-soft-labels-29f3109d9840 – Marco Cerliani Mar 16 '21 at 13:24

0 Answers0