1
$\begingroup$

I am working on an inventory management system where I have daily/monthly/yearly consumption history for a particular item, which may or may not follow a repeating trend. In order to forecast demands within an upcoming time-frame, which model / distributions are recommended?

Data Format: Data is a series of (x, y) pairs where x is a timestamp and y is count of consumption.

As far as I studied, random forest or tree based models do not handle trend patterns effectively, while linear models do not seem to be appropriate either given the fact that data do not seem to be following a linear pattern.

Is this problem solvable using a single model? Should Linear Regression models be used for this problem?

$\endgroup$
1
  • $\begingroup$ It depends on your data, could you plot of your data, and post the graph? but you can use Autoregressive models, such as arima or (since we are talking about stock which is changes with sales) VAR models could also be an appropriate type of model. Depends on your data but you can do better than linear regression. $\endgroup$ Nov 13 '18 at 15:20
3
$\begingroup$

The model you are looking for is called ARIMA, which are time series models where you can obtain trends, cycles, etc.

The ARIMA models lets you model your univariate data by detecting monthly/annual cycles, natural growth (trends) for a variable depending on the history of the same variable.

$\endgroup$
1
  • $\begingroup$ I posted the question while ago and later on we implemented ARIMA as well with seasonality span that suited our needs. Nonetheless, I will accept your answer as that was the right model to work with. $\endgroup$ Apr 14 '19 at 9:07
1
$\begingroup$

Linear models (such as linear regression) should get the job done, but if you are looking for more sophisticated models either you can add more complexity in regression model or go with deep-learning model with few layers.

$\endgroup$

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.