Tips & Tricks
Choosing a Project
Importing Data
Configure Project
Mining Data
Training BackProp
Training SFAM
Making Predictions
Exporting Results
|
Mining Data
Some of the most valuable uses for NeuNet Pro are data mining, data cleaning and anomaly detection.
The purpose is to discover which records do not fit the pattern of the majority.
These suspicious records may then be viewed, manually marked and exported to a separate list.
How to Use NeuNet Pro for Data Mining
- Configure your project so that the entire database is used for both training and testing.
- Press GO to train for a few cycles through the data, then press STOP to terminate the training early
while an interesting amount of error still remains. At this point, the "easy-to-learn"
records will have the correct predictions, while the anomalies will have incorrect predictions.
- Press Finish on the training screen and you will be delivered to the Browse Tests Results where
you will be able to compare the predicted versus actual for every record in your test set.
- For Backprop, right click on the Difference column header and sort the data descending by difference.
The most anomalous records will float to the top of the list.
Alternatively, you may double click on any point of the scatter graph to see your test data
sorted by closeness to the click point.
- For SFAM, right click on the prediction column header and sort the data.
The incorrect predictions will be shown along with the correct predictions for each class.
Alternatively, you may double click on any cell of the confusion matrix to see the test data
that contributed to that cell.
- Manually browse through the data, using the checkmark column to set a checkmark on
those anomalies you wish to export.
- Use File_Export to export the checkmarked records to your audit list.
|
A Complete Neural Network Development System
|