1 / 15

Statistical Assessment of Event Predictors

Statistical Assessment of Event Predictors. Björn Schelter. Statistical Assessment of Event Predictors and Probabilistic Forecasting. Björn Schelter Andreas Schulze-Bonhage, Hinnerk Feldwisch, Michael Jachan, Jens Timmer, Klaus Lehnertz, Ralph Andrzejak, Florian Mormann. The guidelines.

kylene
Download Presentation

Statistical Assessment of Event Predictors

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. IWSP 4, Kansas City Statistical Assessment of Event Predictors Björn Schelter

  2. IWSP 4, Kansas City Statistical Assessment of Event Predictors and Probabilistic Forecasting Björn Schelter Andreas Schulze-Bonhage, Hinnerk Feldwisch, Michael Jachan, Jens Timmer, Klaus Lehnertz, Ralph Andrzejak, Florian Mormann

  3. IWSP 4, Kansas City The guidelines • Use long-term EEG data without pre-selection • Report results for training and testing data • Provide both sensitivity as well as specificity (time under false warning?) • Statistically validate your results

  4. IWSP 4, Kansas City Comparison Difference vanishes for Poisson distributed seizures.

  5. IWSP 4, Kansas City Results • Analytical significance level • Tests statistical significance – Poisson process • Monte-Carlo based technqiues • Can test statistical significance • Can test for various properties of a given predictor • Powerful if designed correctly and in its asymptotic

  6. IWSP 4, Kansas City What is a true prediction?

  7. IWSP 4, Kansas City Standard approach

  8. IWSP 4, Kansas City Re-raising alarms

  9. IWSP 4, Kansas City Pros and cons • Standard approach can be assessed statistically • Re-raising alarms can be handled statistically BUT Sensitivity of the random predictor is 100%

  10. IWSP 4, Kansas City Solution (?) for re-raising alarms • By Snyder et al.: • Limit the time under warning • Statistics suggested similar to SPC statistics • But • Time under true warning might be extremely long

  11. IWSP 4, Kansas City General idea

  12. IWSP 4, Kansas City t-1 tt+1 Probabilistic Features • Transform features into probability by logistic regression

  13. IWSP 4, Kansas City 0.25: indecisive 50% predictor constant-zero predictor natural predictor 0: perfect predictor Prediction Performance: The Brier Score • Sensitivity is not an appropriate measure for performance here … • Range of Brier score: ... Indicator of seizure occurrence (0/1) ... Brier score Question: When can an estimated Brier scorebe regarded as significant ?(i.e. prediction performance above chance level) [Brier, Monthly Weather Review 78, 1950]

  14. IWSP 4, Kansas City Results: Brier-Scores • Correction for multiple testing: 5 tests MPC: mean phase coherence DSI: dynamical similarity index • Significant results obtained for 3/5 patients [Andrzejak et al., Phys. Rev. E, 67, 2003]

  15. IWSP 4, Kansas City The Group • Andreas Schulze-Bonhage • Armin Brandt • Caronlin Gierschner • Jens Timmer • Hinnerk Feldwisch • Michael Jachan • Raimar Sandner

More Related