Coming from a TM1 background (more business than statistics), it is easy to get stated with modeling once you determine your modeling objective, and Modeler can help with that. IBM SPSS Modeler offers an intuitive interface that will appeal to a wide range of users from the non-technical business user to the statistician, data miner or data scientist.
IBM SPSS Modeler uses “node definitions” to easily add objects to modeling streams. These nodes include data sources, record and field operations, graphing, output/export and of course, modeling. These many modeling nodes (found on the Modeling palette) can be classified depending on what your modeling objective is.
SPSS categorizes modeling objectives into three main – Classification, Segmentation and Association.
With a classification model, you are trying to predict a field, using one or more predictors. Examples would include trying to predict which telecommunications customers are liable to drop their plan and go to another provider and in banking, predicting if a customer might fail on paying back a loan.
With a Segmentation model, grouping records (using one or more fields) is the idea. For example a marketing group may “cluster” customers based upon RFM (Recency, Frequency and Monetary values) and insurance companies may cluster claims and look for unusual cases within the groups to detect fraud.
Association models look for relationships between fields to try to find the “rules of the format”, for example: a certain percentage of customers have purchased both products “A” and product “B” and those customers also have purchased product “C”.
Once you determine which type of model you ar interested in you are ready to being modeling (with the assistance of SPSS).
The SPSS Modeling palette can be organized by selecting one of the three objectives. When you make a selection, Modeler will “suggest” the modeling nodes that are applicable to that selection – which is great for us TM1 guys new to modeling with SPSS Modeler!
Once you have determined where your objective “fits” you’ll see that more than just one model “type”
can be used. Keep in mind that the business context (objective) will be the first decider in the choice of a model type, but other factors will also influence the decision such as:
- How missing values should be handled
- How categorical predictors should be handled
- How continuous predictors should be handled
- How will a particular model score data?
But in the end, it is always the business user, balancing all pros and cons, who will decide which model, should be used.