Schedule

Changed by: Tom Fawcett, 23-Jul-1997

8:30 -- 8:50	Introductory Remarks
8:50 -- 9:20	Prospective Assessment of AI Technologies for Fraud Detection: A Case Study (Jensen)
9:25 - 9:55	Break Detection Systems (Goldberg and Senator)
10:00 - 10:25	Coffee Break
	Technical Session 1
10:25 - 10:50	Detecting Cellular Fraud Using Adaptive Prototypes (Burge and Shawe-Taylor)
10:50 - 11:15	Sequence Matching and Learning in Anomaly Detection for Computer Security (Lane and Brodley)
11:15 - 11:40	Intrusion Detection with Neural Networks (Ryan, Lin and Miikkulainen)
11:40 - 12:00	Discussion
12:00 - 1:00	LUNCH
	Technical Session 2
1:00 - 1:25	A Multi-Agent Systems Approach for Fraud Detection in Personal Communication Systems (Abu-Hakima, Toloo and White)
1:25 - 1:50	Clustering and Prediction for Credit Line Optimization (Haimowitz and Schwarz)
1:50 - 2:15	Analysis and Visualization of Classifier Performance with Nonuniform Class and Cost Distributions (Provost and Fawcett)
2:15 - 2:30	BREAK
	Technical Session 3
2:30 - 2:55	JAM: Java Agents for Meta-Learning over Distributed Databases (Stolfo, Prodromidis, Tselepis, Lee, Fan and Chan)
2:55 - 3:20	Credit Card Fraud Detection Using Meta-Learning: Issues and Initial Results (Stolfo, Fan, Lee, Prodromidis and Chan)
3:20 - 4:10	Summary/Synthesis (Provost)
4:10 - 4:30	Coffee Break
4:30 - 5:10	Panel: Detecting Consumer Risk and Fraud: Privacy versus Accuracy and Payoff
5:10 - 5:30	Discussion
5:30 - 5:45	Wrap-up

Notes:

This schedule is more-or-less final and is unlikely to change.
"Coffee Break" means a AAAI-specified break (with coffee, pastries, etc.) which all the workshops take together. A "BREAK" is just for this workshop.)

Panel

Detecting Consumer Risk and Fraud: Privacy versus Accuracy and Payoff

As we in the corporate world know, the more information we have on consumers, the more accurate our predictions can be of bad credit risks, and potential fraud. However, not all information is legal or ethical to collect and use. Please try to answer the following questions as they relate to your work:

Do businesses model via automated systems the entire customer population? or do they model at the individual customer profile level?
What are the tradeoffs in "accuracy" of prediction in either case?
What are the potential liabilities inherent in the latter case? If there is improved predictive accuracy, is it worth the potential liabilities?
Are individual user models linked or integrated with other information, and used for other business applications (eg, marketing)?
Might the absence of certain unusable, statutory information significantly reduce model accuracy?
What technological solutions are needed to ensure both customer privacy, and the real needs of business to reduce its losses and stem the tide of fraud and written-off debt?
How can we use Artificial Intelligence (e.g. knowledge-based) approaches to both model and to insure compliance with the ethical/legal restrictions.

Suggested reading is the NY Times article on Front page and B14 of Thursday June 12,1997, entitled On Frontier of Cyberspace, Data is Money, and a Threat.

Tom Fawcett <fawcett@nynexst.com>

Changed by: Tom Fawcett, 23-Jul-1997