Download
application 2 misstatement detection n.
Skip this Video
Loading SlideShow in 5 Seconds..
Application 2: Misstatement detection PowerPoint Presentation
Download Presentation
Application 2: Misstatement detection

Application 2: Misstatement detection

66 Views Download Presentation
Download Presentation

Application 2: Misstatement detection

- - - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript

  1. Application 2: Misstatement detection • Problem: Given network and noisy domain knowledge about suspicious nodes (flags), which nodes are most risky? Transfer from A/R, report revenue Money from A/P to purchase inventory Inventory Revenue Orange Cty Accounts Payable Cash Revenue Los Angeles Bad Debt Accounts Receivable Revenue San Diego Revenue San Francisco Non-Trade A/R Revenue Other

  2. Application 2: Misstatement detection • Problem: Given network and noisy domain knowledge about suspicious nodes (flags), which nodes are most risky? Inventory Revenue Orange Cty Round-dollar entries Accounts Payable Cash Revenue Los Angeles Bad Debt Accounts Receivable Revenue San Diego Large number of returns Many late postings Revenue San Francisco Non-Trade A/R Revenue Other Many entries reversed late in period

  3. Application 2: Misstatement detection • Solution: Social Network Analytic Risk Evaluation • Assume homophilybetween nodes (“guilt by association”) • Use belief propagation (message passing) • Upon convergence, determine end risk scores. • Details: See Ch. 9.3.

  4. Application 2: Misstatement detection • Nodes in proximity of many flags will be marked as risky, nodes flagged in isolation will not. Inventory Revenue Orange Cty Accounts Payable Cash Revenue Los Angeles Bad Debt Accounts Receivable Revenue San Diego Revenue San Francisco Non-Trade A/R Revenue Other

  5. Application 2: Misstatement detection • Nodes in proximity of many flags will be marked as risky, nodes flagged in isolation will not. Focus on staff posting to A/R from headquarters Inventory Revenue Orange Cty Accounts Payable Cash Revenue Los Angeles Bad Debt Accounts Receivable Revenue San Diego Ignore A/P, no corroborating evidence Revenue San Francisco Non-Trade A/R Revenue Other

  6. Application 2: Misstatement detection • Accurate- up to 6.5 lift • Flexible- Can be applied to other domains • Scalable- Linear time • Robust- Works on large range of parameters Results for accounts data (ROC Curve) Ideal SNARE True positive rate Baseline (flags only) False positive rate