Redigerer Mal:Forvirringsmatrisebegreper

{| class="wikitable" width=35% style="float:right;font-size:98%; margin-left:0.5em; padding:0.25em; background:#f1f5fc;"
|+ Terminologi og utledninger<br />innen forvirringsmatriser
|- style="vertical-align:top;"
|
; tilstand positiv (P)
:antall reellt positive tilfeller i datene
; tilstand negativ (N)
:antall reellt negative tilfeller i datene
----
; sann positiv (SP)
:antall reellt positive tilfeller klassifisert som positive (trefninger)
; sann negative (SN)
:antall reellt negative tilfeller klassifisert som negative (korrekt forkastninger)
; falsk positive (FP)
:antall reellt negative tilfeller klassifisert som positive (falske alarmer, type-I-feil)
; falsk negative (FN)
:antall reellt positive tilfeller klassifisert som negative (bomskudd, type-II-feil)
----
; [[Sensitivitet|sensitivitet]], ''recall'', trefningsrate, eller sann-positiv-rate (SPR)
:<math>\mathrm{SPR} = \frac {\mathrm{SP}} {\mathrm{P}} = \frac {\mathrm{SP}} {\mathrm{SP}+\mathrm{FN}}= 1 - \mathrm{FNR}</math>
; [[Spesifisitet|spesifisitet]], selektivitet, eller sann-negativ-rate (SNR)
:<math>\mathrm{SNR} = \frac {\mathrm{SN}} {\mathrm{N}} = \frac {\mathrm{SN}} {\mathrm{SN} + \mathrm{FP}} = 1 - \mathrm{FPR}</math>
; presisjon, eller positiv-prediktiv-verdi (PPV)
:<math>\mathrm{PPV} = \frac {\mathrm{SP}} {\mathrm{SP} + \mathrm{FP}} = 1 - \mathrm{FOR}</math>
; negativ-prediktiv-verdi (NPV)
:<math>\mathrm{NPV} = \frac {\mathrm{SN}} {\mathrm{SN} + \mathrm{FN}} = 1 - \mathrm{FOR}</math>
; bomskuddsrate, eller falsk-negativ-rate (FNR)
:<math>\mathrm{FNR} = \frac {\mathrm{FN}} {\mathrm{P}} = \frac {\mathrm{FN}} {\mathrm{FN} + \mathrm{SP}} = 1 - \mathrm{SPR} </math>
; ''fall-out'', eller falsk-positiv-rate (FPR)
:<math>\mathrm{FPR} = \frac {\mathrm{FP}} {\mathrm{N}} = \frac {\mathrm{FP}} {\mathrm{FP} + \mathrm{SN}} = 1 - \mathrm{SNR}</math>
; falsk-oppdagelsesrate (FOR)
:<math>\mathrm{FOR} = \frac {\mathrm{FP}} {\mathrm{FP} + \mathrm{SP}} = 1 - \mathrm{PPV} </math>
; falsk-utelatelsesrate (FUR)
:<math>\mathrm{FUR} = \frac {\mathrm{FN}} {\mathrm{FN} + \mathrm{SN}} = 1 - \mathrm{NPV} </math>
----
; nøyaktighet (NØY)
:<math>\text{NØY} = \frac {\mathrm{SP} + \mathrm{SN}} {\mathrm{P} + \mathrm{N}} = \frac {\mathrm{SP} + \mathrm{SN}} {\mathrm{SP} + \mathrm{SN} + \mathrm{FP} + \mathrm{FN}} </math>
;[[F1-verdi]]
: det harmoniske gjennomsnitt mellom presisjon og sensitivitet
:<math>\mathrm{F}_1 = 2 \cdot \frac {\mathrm{PPV} \cdot \mathrm{SPR}} {\mathrm{PPV} + \mathrm{SPR}} = \frac {2 \mathrm{SP}} {2 \mathrm{SP} + \mathrm{FP} + \mathrm{FN}}</math>
; Matthews korrelasjonskoeffisient (MKK)
:<math> \mathrm{MCC} = \frac{ \mathrm{SP} \times \mathrm{SN} - \mathrm{FP} \times \mathrm{FN} } {\sqrt{ (\mathrm{SP}+\mathrm{FP}) ( \mathrm{SP} + \mathrm{FN} ) ( \mathrm{SN} + \mathrm{FP} ) ( \mathrm{SN} + \mathrm{FN} ) } }</math>
; Youden-indeks, ''Informedness'', eller ''Bookmaker Informedness'' (BM)
:<math> \mathrm{BM} = \mathrm{SPR} + \mathrm{SNR} - 1</math>
; Markerthet, ''Markedness'' (MH)
:<math> \mathrm{MK} = \mathrm{PPV} + \mathrm{NPV} - 1</math>

<span style="font-size:90%;">''Kilder: Fawcett (2006),<ref name=Fawcett2006>{{cite journal|last=Fawcett|first=Tom|title=An Introduction to ROC Analysis|journal=Pattern Recognition Letters|date=2006|volume=27|issue=8|pages=861–874|doi=10.1016/j.patrec.2005.10.010|url=http://people.inf.elte.hu/kiss/11dwhdm/roc.pdf}}</ref> Powers (2011),<ref name="Powers2011">{{cite journal |first=David M W |last=Powers |date=2011 |title=Evaluation: From Precision, Recall and F-Measure  to ROC, Informedness, Markedness & Correlation |journal=Journal of Machine Learning Technologies |volume=2 |issue=1 |pages=37–63 |url=http://www.flinders.edu.au/science_engineering/fms/School-CSEM/publications/tech_reps-research_artfcts/TRRA_2007.pdf}}</ref> and Ting (2011)<ref name=ting2011precision>{{cite book |last=Ting |first=Kai Ming |title=Encyclopedia of machine learning |date=2011 |publisher=Springer |isbn=978-0-387-30164-8 |url=https://link.springer.com/referencework/10.1007%2F978-0-387-30164-8 }}</ref>''
</span>
|}