CODEWITHSUNDEEP
python-3.xmachine-learningtime-seriesforecastingarima
Venugopal srinivasan
Venugopal srinivasan

Reputation: 145

Arima with multivariate independent variables in python

I have a dataset having dateofpurchase,locations,items,salesqty as shown below,

Date        Location    Item    sales_qty
02/01/2019    aaa        x        123
02/01/2019    aaa        y        323
02/01/2019    bbb        x        1023
02/01/2019    bbb        y        1203

I have this type of data for 2 years,25 different locations,400 different item set.I want to forecast my sales on all the locations and item level.I'm new to the time series with multivariate data.Please help me to forecast or give some ideas to me.Thanks in advance.

Upvotes: 4

Views: 5540

Answers (1)

C8H10N4O2
C8H10N4O2

Reputation: 19005

In the statsmodels module, the class statsmodels.tsa.statespace.varmax.VARMAX is likely your best option.

Vector Autoregressive Moving Average with eXogenous regressors model

Notice there's no I (differencing) component, so you will have to ensure stationarity beforehand. (Use statsmodels.tsa.stattools.adfuller and kpss). Also, you will need to deterime the order (p,q) of your ARMA in advance. (Use statsmodels.tsa.stattools.acf and pacf to do this.)

There's also the module statsmodels.vector_ar, which supports supports only AR (not MA) components. You can specify the number of AR terms in the fit method, but by default it does lag selection.

Upvotes: 4

Related Questions