算法評價指標 - FAR,FRR,ERR (TP/FP/FN/TN)

TP/FP/FN/TN

• True positive: having the condition and tested positive with the condition
• False positive: not having the condition but tested positive with the condition
• True negative: not having the condition and tested negative with the condition
• False negative: having the condition but tested negative with the condition

• Sensitivity (True positive rate, TPR): measures the proportion of positives that are correctly identified
• Specificity (True negative rate,TNR): measures the proportion of negatives that are correctly identified

FAR, FRR, ERR

FAR

FAR (False Accept Rate) refers to the possibility that an imposter is accepted by the system incorrectly, which may lead to security breach.

FAR(誤接受率)指冒名頂替者被系統錯誤接受的可能性，這可能致使安全漏洞.

FRR

The FRR (False Reject Rate) is the possibility that a legitimate user is rejected by the system when requesting protected resources.

FRR(誤拒絕率)是合法用戶請求受保護資源時被系統拒絕的可能性。

ERR

在理想狀況下，FAR和FRR均爲零。可是在現實中這是不可能的，由於這兩種偏差是相互矛盾的，一種偏差的減小會致使另外一種偏差的增長。所以，須要用EER (Equal Error Rate)來評估擊鍵動力學認證系統的總體性能。

ERR是FRR等於FAR時的錯誤率，能夠將FAR和FRR統一爲一個參數，更好地表明系統的綜合性能。通常來講，ERR越低，識別系統的性能越好。