BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Isaac Newton Institute Seminar Series
SUMMARY:Assessing Evidentiary Value in Fire Debris Analysi
s - Michael Sigman (University of Central Florid
a)
DTSTART;TZID=Europe/London:20161110T153000
DTEND;TZID=Europe/London:20161110T161500
UID:TALK68926AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/68926
DESCRIPTION:Co-author: Mary R. Williams (National Center
for Forensic Science\, University of Central Flo
rida)
This presentation will
examine the calculation of a likelihood ratio to
assess the evidentiary value of fire debris analy
sis results. Models based on support vector machi
ne (SVM)\, linear and quadratic discriminant analy
sis (LDA and QDA) and k-nearest neighbors (kNN) m
ethods were examined for binary classification of
fire debris samples as positive or negative for i
gnitable liquid residue (ILR). Computational mixi
ng of data from ignitable liquid and substrate py
rolysis databases was used to generate training an
d cross validation samples. A second validation w
as performed on fire debris data from large-scale
research burns\, for which the ground truth (posi
tive or negative for ILR) was assigned by an anal
yst with access to the gas chromatography-mass spe
ctrometry data for the ignitable liquid used in t
he burn. The probabilities of class membership we
re calculated using an uninformative prior and a l
ikelihood ratio was calculated from the resulting
class membership probabilities . The SVM method
demonstrated a high discrimination\, low error rat
e and good calibration for the cross-validation d
ata\; however\, the performance decreased signific
antly for the fire debris validation data\, as in
dicated by a significant decrease in the area und
er the receiver operating characteristic (ROC) cur
ve. The QDA and kNN methods showed performance tr
ends similar to those of SVM. The LDA method gave
poorer discrimination\, higher error rates and sl
ightly poorer calibration for the cross validatio
n data\; however the performance did not deteriora
te for the fire debris validation data.
LOCATION:Seminar Room 1\, Newton Institute
CONTACT:INI IT
END:VEVENT
END:VCALENDAR