3 Replies Latest reply: May 23, 2012 4:28 PM by Charlie Berger-Oracle RSS

    Using classification for regression purposes

    920802
      Hi,


      How configure ODM to make a regression using multiple target value capabilities of the classification model ?
      I have dataset with "month_number" and "year_number" [number] (3 years), "prod_id" [number] and "value" [number] and multiple prod_id within it. I want to predict "value" for few next months for this multiple prod_id.
      Is it possible using one classification node in workflow?
      What should be a case_id? Or I should not set case_id ? When I assign prod_id as case_id it generates error that it does not contain unique values.
      If I leave case id as none model is generated, by predicted values....
      I think are not right.... The same values are assigned to months for all prod_id's. It means that there is the same value for january in all prod_id, other value is for february - but the same in all prod_id, etc.

      I train model using past data. For prediction I prepared dataset with further dates (+1 to +6 months), set proper prod_id and assigned "value" as 0 (or null - result is the same) . Is it correct ?


      Best regards,
      Paul.

      Edited by: 917799 on 2012-05-22 03:49

      Edited by: 917799 on 2012-05-22 03:57

      Edited by: 917799 on 2012-05-22 04:00

      Edited by: 917799 on 2012-05-22 06:43
        • 1. Re: Using classification for regression purposes
          Charlie Berger-Oracle
          What you describe sounds more like a forecast than a prediction. Time series forecasting typically involves looking at product revenues or amounts sold by month over some long period of time to them make a forecast of amounts sold in the upcoming N months. Time Series generally focuses on aggregate level forecasts over a series of future time periods whereas ODM is more about making a prediction and is more focused on predicting an individual;s behavior e.g. Who will buy a Buick in the near future? You can sum all the predictions up and get a forecast for month N1 but this isn't really what you are looking for I think. I think you might want to read this blog re: How to perform Time Series forecasting using ODM's SVM algorithm http://oracledmt.blogspot.com/2006/01/time-series-forecasting-part-1_23.html
          • 2. Re: Using classification for regression purposes
            920802
            Yes, it's forecast.

            Some time ago I posted a question on: Multiple Regression in ODM

            I received advice: "you can either build a single model and use the multiple target value capabilities of the classification model to handle this use case. The Build node incorporates a "balanced" performance setting, that does a pretty good job at insuring an high average accuracy".

            But I don't know how technically use classification (how to configure model in ODM?) for making forecast for multiple product id's (for only +1 month as I undestand).

            Regarding time series and article: this article relates to use PL/SQL and my question is whether it is possible to make such forecasts for many prod id and for many further months using only workflow.
            • 3. Re: Using classification for regression purposes
              Charlie Berger-Oracle
              Have you looked at the e-training on ODM or signed up for the Oracle Virtual Training course? I think that would help you. See link to Getting Started w/ ODM and related blog entrys:

              https://blogs.oracle.com/datamining/entry/evaluating_oracle_data_mining_has

              cb