Is it possible to adapt algorithms (after data pre-processing)?

[Jonasvr]

I have to write a paper on data pre-processing and its effects on the cost sensitivity of models. Being new to SAS I am wondering if it is possible to adapt/change the algorithms that I will be using. Thanks in advance.

[ballardw]

If I have any clue as to what you may be getting at I think you are asking about comparing the results of processing data in two or more different ways. Generally create one set of output data from each approach and then compare the final results. Depending on the type of algorithm you need it may be easy or difficult.

Between the SAS/IML (Interactive Matrix Language) and datastep there are many tools for raw data manipulation. A common challenge for new SAS users is finding out when to drop out of the low-level manipulation provided by IML and the data step and move to summary procedures or other statistical procedures.

[Jonasvr]

I was not exactly sure how to ask, allow me to rephrase.. Is it possible to create a new classifier in SAS by modifying the internal processes of an existing method, such as Random Forests or Adaboost for instance?

[ballardw]

If you mean creating a new option for an existing procedure then a qualified "no". Sometimes you can accomplish similar effects by creation of additional variables and combinations of existing features but user extentions generally aren't possible. However the SAS/IML matrix language can be used to program many of the modeling procedures and allow the introduction of your algorithms.

You may want to post a question with a few more details in either the SAS/IML Software or Mathematical Optimization, Discrete Event Simulation for general modeling or SAS Forecasting and Econometrics (if your need is time-series related) to get the attention of folks more involved in the guts of modeling.