BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:Dissertation Proposal: Laura Bond
X-WR-TIMEZONE:Mountain Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20261014T223351Z
UID:tag:localist.com\,2008:EventInstance_53365374544752
DTSTART:20260716T143000Z
DTEND:20260716T153000Z
DESCRIPTION:Dissertation Information\n\nTitle: Prediction-Focused Cross-Val
 idation for Block-Clustered Data\n\nProgram: Computing Ph.D. Data Science 
 emphasis\n\nAdvisor: Dr. Grady Wright\n\nCommittee Members: Dr. Edoardo Se
 rra\, Dr. Michael Perlmutter\n\nAbstract:\n\nCross-validation remains the 
 default tool for estimating prediction error and selecting models for pred
 icting observations in a data set.  Block-clustered data\, that is\, obser
 vations nested within larger groups such as patients in clinics or student
 s in schools\, are common\, and violate the independence that standard k-f
 old cross-validation assumes.  When the goal is prediction to new clusters
 \, k-fold cross-validation is optimistic due to the leakage of within-clus
 ter information into the folds\, from the within-cluster correlation of ob
 servations and the cluster structure being mistaken for signal.  Theoretic
 al work has demonstrated that prediction to observations within existing c
 lusters does not optimistically bias the k-fold cross-validation predictio
 n error\, and that leave-one-cluster-out (LOCO) cross-validation targets t
 he true generalization error.  \n\nThe proposed thesis addresses open ques
 tions regarding when the LOCO – 10-fold gap is large enough to affect de
 cisions\, and when it might practically be ignored (e.g. if the within-clu
 ster correlation is low\, or if the overall sample size is sufficiently la
 rge).  Proposed simulation studies address these questions for cluster-agn
 ostic and cluster-aware machine learning methods for continuous and binary
  outcomes\, covering parametric models with known corrections to non-linea
 r ensemble methods without these corrections.  \n\nFor the latter simulati
 ons\, the generalization error will be built as ground truth from unseen c
 lusters.  How precisely the LOCO estimate itself can be reported in the pa
 rametric case is also studied.  Controlled data-generating mechanisms span
 ning a range of intraclass correlations\, cluster sizes\, and total sample
  sizes are used\, with methods judged on the estimated gap size and wall t
 ime.  It is expected that these studies will provide practical guidance on
  when cluster-aware validation is necessary\, relative to standard k-fold 
 calculations\, specific to the data in-hand and the intended prediction ap
 proaches.
GEO:43.604437;-116.205673
LOCATION:Mathematics Building (MB)\, 205
SUMMARY:Dissertation Proposal: Laura Bond
URL;VALUE=URI:https://events.boisestate.edu/event/dissertation-proposal-lau
 ra-bond
CATEGORIES:Lectures and Presentations
END:VEVENT
END:VCALENDAR