I have a doubt in below ML families.

If we are predicting: Yes, then we have classification and regression

If No, then we have clustering

In clustering, we have K-means algo

In classification we predict discrete values, so we have logistic regression, KNN, and NN

In Regression we predict continuous values, so we have Simple Linear and Multiple Linear regression, also called as causal models

Then we have TimeSeries, with Single, Double and Triple exponential models. And ARIMA, SARIMA models

ARIMA combines Regression and Simple means(Moving Average). Now my question is, does TimeSeries come under classification or Regression?

  • $\begingroup$ You present what looks to me to be a false dichotomy. I think part of the difficulty may be arising from the tendency of machine learning to treat all problems as one of clustering, classification or prediction. The scope of statistical analyses is considerably broader than those applications. An analysis of time series data may be for neither classification nor prediction (let alone actually be regression). A variety of time-series models are akin to regression models (though the ones that are are not necessarily being used for prediction). However some time series models would ...ctd $\endgroup$ – Glen_b -Reinstate Monica Nov 3 '18 at 7:28
  • $\begingroup$ ctd... considerably stretch the notion of regression. Why seek to pound a somewhat roundish peg into an unneccesarily square hole? $\endgroup$ – Glen_b -Reinstate Monica Nov 3 '18 at 7:28

Strictly speaking, time series methods don't correspond to a specific family of ML models like classification or regression etc... They are more of a domain of application, similar to natural language processing, image processing, etc...

Models from the Exponential Smoothing family are similar in spirit to regression models, however they are different from regression top problems in the sense that they are sequential and take a variable (possibly infinite number of inputs). You can see this by expanding the expression of an exponential smoothing model.

ARIMA models are an interesting case: If they have only an AR component (meaning ARIMA(p,d,q) with q=0) then they are regression models (AR stands for Auto-Regression after all) - however if the have an MA component, then they fall in the same category as Exponential Smoothing models.

See this post for what ES are not considered regression models and this post for why the MA component of ARIMA models makes them different from regression models.


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.