Business Context
A manufacturing plant tracks defects per batch. However, sensors often fail, leading to missing values (NULLs) in the temperature and pressure readings. The team wants to use LASSO selection to find the most critical factors despite the data quality issues.
Data Preparation
Creation of a dataset with random missing values (NULL) in predictor variables to test robustness.
Copied!
| 1 | |
| 2 | DATA mycas.mfg_defects; |
| 3 | call streaminit(456); |
| 4 | DO batch = 1 to 500; |
| 5 | Temperature = rand('NORMAL', 100, 10); |
| 6 | IF rand('UNIFORM') < 0.1 THEN Temperature = .; |
| 7 | Pressure = rand('NORMAL', 50, 5); |
| 8 | IF rand('UNIFORM') < 0.05 THEN Pressure = .; |
| 9 | Defects = rand('POISSON', exp(0.02 * (ifn(Temperature=., 100, Temperature)) - 0.01 * Pressure)); |
| 10 | OUTPUT; |
| 11 | END; |
| 12 | |
| 13 | RUN; |
| 14 |
Étapes de réalisation
1
Attempt variable selection (LASSO) with missing values in regressors and save the model.
Copied!
| 1 | |
| 2 | PROC CAS; |
| 3 | countreg.countregFitModel / TABLE={name='mfg_defects'} model={depVars={{name='Defects'}}, effects={{vars={'Temperature', 'Pressure'}}}, modeloptions={modeltype='POISSON'}} selection={method='LASSO'} store={name='defect_model_store', replace=true}; |
| 4 | |
| 5 | RUN; |
| 6 |
2
Verify that the model store was created despite the missing values (observations with missing values should be dropped or handled).
Copied!
| 1 | |
| 2 | PROC CAS; |
| 3 | TABLE.fileInfo / path='defect_model_store.sashdat'; |
| 4 | |
| 5 | RUN; |
| 6 |
Expected Result
The action completes successfully. Observations with missing values in 'Temperature' or 'Pressure' are excluded from the analysis (standard SAS behavior), but the process does not crash. The 'defect_model_store' is successfully created for future scoring.