Spam Filter Project Update
Speakers: Machine-Learning Data Analysts: Harsha Gopianandan, Ananth Gopalakrishnan
Despite spikes in spam over the six weeks or so, the performance of the spam filter service is netting a 95-98% accuracy. Below are the stats from Nov 23 - 27, 2019.
November 23 - 27
Total applications: 19,943
True Positive 938
True Negative 18,965
False Positive 13
False Negative 22
Accuracy 99.7994283709 %
Precision 98.6330178759 %
Recall 97.7083333333 %
Notes from Nov 12 meeting:
Harsha Gopianandan from the Spam Filter Machine Learning team explained about false positives and false negatives and why we are seeing both across the system.
He explained how the model works and encouraged colleges to keep tagging diligently so that model can learn from all new signatures.
Getting ready to update the model again with the Model with PII (December 2019).
We talked about sending false negatives to the Tech Center and we will bulk tag them and upload to the model, so the model can benefit from the identified false negatives
- August 24 - Last Model Update
- November 22 - Pii Data Added to Model and Other Enhancements (IP Region, Email Domain)
- December 12 - Manual Retraining Model with Updated Data
Update the model every 2-3 weeks.
Update with the latest data.
Update the model with new features.
*Note: Derived data from PII data is used.
Taking that info and transforming it in a way that is useful to the model.
Name + DOB
Are you seeing a large amount of False Negatives?
If your college is seeing blocks of false negatives (fraud applications that were NOT caught and suspended by the spam filter)...we can help you tag them so that all colleges benefit from those signatures,
Please send quantities of 20 or more to Support Services using the instructions below.
SPAM Drop File Information
Please provide bulk fraudulent applications in the format specified below:
- File Format = .TXT
- File Naming Convention = CollegeMISCode_Fraud_mmddyy.txt
- Confirmation # (only1 confirmation # per line)
Send to: email@example.com
We ask that all colleges follow the file format above. If you would like to include any information other than confirmation #, please provide that information in a separate file.
For other assistance, please contact:
Merrie Wales firstname.lastname@example.org
Patty Donohue email@example.com