Presentation is loading. Please wait.

Presentation is loading. Please wait.

Discriminant Analysis

Similar presentations


Presentation on theme: "Discriminant Analysis"— Presentation transcript:

1 Discriminant Analysis

2 Overview Used for Categorical dependent variable.
Goal is to classify records into 2 or more groups. When only 2 groups for dependent, it is equivalent to a multiple linear regression model.

3 Basic Process Building model Evaluation
Break down independent variables into dummies (continuous variables can be used also). Perform regression with the dummy dependent (assuming a two-valued dependent). Use multiple iterations of regression until variables are significant and meaningful in the context of the model. Evaluation Score the data – both training and validation data. Perform K-S test.

4 SAS Program – Creating Dummies
Data save.train2; Set save.train; if dwnpmt >= 1000 and dwnpmt < 2000 then dwnpmt1 = 1; else dwnpmt1 = 0; if dwnpmt >= 2000 then dwnpmt2 = 1; else dwnpmt2 = 0; And so on for other variables …. run;

5 SAS Program - Regression
proc reg data=save.train2 corr outest=estfile; bgscore: model good = trades01 trades02 ageavg01 ageavg02 ageavg03 …. List all dummies / selection=none; Run;

6 Scoring the dataset proc score data=save.train2 score=estfile out=save.scrdata type=parms; var trades01 trades02 ageavg01 ageavg02 ageavg03 …. Same variables as in the last model run run; Do the same for the save.valid2 dataset, after first running the dummy program to create the same dummies for the validation dataset as in the training dataset.

7 Score Crosstabs Format the score variable (bgscore) into 50 or 100 point intervals, and create crosstab against the dependent variable (Good). PROC DATASETS LIBRARY=SAVE; MODIFY SCRDATA; FORMAT bgscore insert_format_name_here. ; RUN; QUIT; You should see more of the Good category among the higher score bands and more of the Bad category in the lower score bands, if the model is in fact able to discriminate.

8 Evaluate Perform KS test to evaluate the model, and create a forecast.
See K-S spreadsheet


Download ppt "Discriminant Analysis"

Similar presentations


Ads by Google