Business Context
An industrial manufacturing plant wants to predict equipment failure using sensor data. However, due to network issues and sensor malfunctions, the data is often incomplete. This scenario tests the model's robustness and its ability to handle missing values in both interval and nominal variables through imputation.
About the Set : bayesianNetClassifier
Classification using Bayesian networks.
Discover all actions of bayesianNetClassifier Data Preparation
Creation of a sensor dataset with significant missing data. Both interval ('Temperature', 'Pressure') and nominal ('SensorHealth') variables have missing values, represented by '.' for numeric and empty strings for character.
Copied!
| 1 | DATA casuser.sensor_data; |
| 2 | LENGTH Failure_Risk $4. SensorHealth $6.; |
| 3 | INFILE DATALINES delimiter=','; |
| 4 | INPUT MachineID $ Temperature Pressure Vibration SensorHealth $ Failure_Risk $; |
| 5 | DATALINES; |
| 6 | M01,250.5,15.2,0.5,Normal,Low |
| 7 | M02,310.1,.,0.8,Normal,High |
| 8 | M03,245.8,14.9,1.2,,High |
| 9 | M04,.,15.5,0.4,Normal,Low |
| 10 | M05,295.0,19.8,0.7,Alert,High |
| 11 | M06,260.3,16.1,0.5,Normal,Low |
| 12 | M07,330.0,21.0,.,,High |
| 13 | M08,.,.,0.6,Alert,Low |
| 14 | ; |
| 15 | RUN; |
Étapes de réalisation
1
Train a Parent-Child (PC) structure model. Configure the action to impute missing interval variables with the mean and missing nominal variables with the mode.
Copied!
| 1 | PROC CAS; |
| 2 | bayesianNetClassifier.bnet |
| 3 | TABLE={name='sensor_data'} |
| 4 | target='Failure_Risk' |
| 5 | inputs={'Temperature', 'Pressure', 'Vibration', 'SensorHealth'} |
| 6 | nominals={'SensorHealth', 'Failure_Risk'} |
| 7 | structures={'PC'} |
| 8 | missingInt='IMPUTE' |
| 9 | missingNom='IMPUTE' |
| 10 | alpha=0.1 |
| 11 | OUTPUT={casout={name='sensor_scored', replace=true}, copyVars={'MachineID'}} |
| 12 | display={'MissInfo', 'NetInfo'}; |
| 13 | RUN; |
2
Check the 'MissInfo' table in the results to confirm that imputation was applied to the specified variables.
Copied!
| 1 | /* The 'MissInfo' table is displayed directly in the results of the previous step. Manual review of the log is expected. */ |
Expected Result
The action must run without errors, demonstrating its ability to handle a mix of missing data types. The 'MissInfo' table in the output log should confirm that 'Temperature', 'Pressure', and 'SensorHealth' were imputed. A scored output table 'sensor_scored' should be created with predictions for all input rows, including those that originally had missing values. This validates the imputation strategy's effectiveness.