Machine learning algorithms for time series analysis

3

I am wondering which ML algoirthms (supervised) are commonly used for TS analysis. Which ones have used and found to be succesful for your projects?

Robot_enthusiast

Posted 2018-01-31T00:07:01.387

Reputation: 51

Question was closed 2018-02-01T18:12:42.060

1Look into sequence-to-sequence (neural network) models. – Emre – 2018-01-31T00:44:57.930

1Are you interested in a specific subfield of time series analysis, like forecasting for example? – Imran – 2018-01-31T05:31:01.793

Long short term memories may help you – Media – 2018-01-31T06:22:18.670

@Imran indeed I am – Robot_enthusiast – 2018-01-31T10:23:23.150

Answers

0

It mainly depends on the task you are looking to accomplish with your time series data. Is it a regression or a classification task. In general, Random Forests are known to be easy to use and pretty useful. So I think it's a good way to start. Of course, dimensionality reduction techniques and feature extractors like PCA, kernel PCA, t-SNE and many others are very useful if you want to get better results. Neural networks are also useful, depending on the task you can try both recurrent (LSTM) or convolutional NN.

HatemB

Posted 2018-01-31T00:07:01.387

Reputation: 196

1I guess pointing OP to the resources where someone has successfully implemented regression tasks using above techniques seems better. Op can then decide what he/she wants to achieve and what to use. – Kiritee Gak – 2018-01-31T12:31:51.710

what about in the domain of scikit learn? Any techniques that stick out from there which are known ot be useful for time series forecasting? – Robot_enthusiast – 2018-01-31T13:23:32.117

There are a lot of linear models. Check out arima and it's variants. They come in statsmodel library. Pyflux is a new library, fairly, that got some time series models implemented. – Kiritee Gak – 2018-02-01T03:23:45.017

regression task @HatemB – Robot_enthusiast – 2018-02-01T23:06:51.297

Yes in fact, there are a lot of linear autoregressive you can maybe start with like arima unfortunately I don't think that autoregressive models are implemented within scikit learn but you can easily find libraries that provides them on python or R. There are also many non linear predictors like random forests regressor. – HatemB – 2018-02-06T14:12:30.713