I want to have your opinion on the performance of my SVM classifier (k-fold cross validated):

Classification of class1(n=45)/class2(n=86)

Accuracy: 65.4%

Sensitivity: 88.2%

Specificity: 22.2%

AUC: 0.628

How can I interpret the high sensitivity and low specificity?!


you have high true positive rate which is 88.2% that means 88.2% time your model predicted positive classes which are actually positive,example : percentage of cancerous people who are correctly identified as having the condition.it also indicates that your recall score is high,we know that Recall = (tp/(tp+fn)) here tp = true positive and fn= false negative rates

but your specificity is low which indicates that "true negative rates" of your model is low which is 22.2%.example : the percentage of non cancerous patients who are correctly identified as not having the condition it is also known as precision and we know precision = (tp/(tp+fp)) here tp = true positive rates and fp = false positive rates

shortly : *your model has a precision of 22%, when it predicts a tumor is malignant, it is correct 22% of the time.

*your model has a recall of 88%, it correctly identifies 88% of all malignant tumors.

conclusion : perhaps your model is overfitting(you need to check your cross validation score)and your model has low accuracy and high false positive(fp) rates that means your model is making many mistakes and should not do well on unseen data

  • $\begingroup$ Ok so it doesn't mean that my model failed or that it's useless? Also, is it possible that i have this results because of unbalanced classes (45 vs 86)? $\endgroup$ – learneRS May 23 at 15:54
  • $\begingroup$ your model has failed and yeah unbalanced classes often lead to such problems,your model is kind of biased now,you need to improve it,thanks $\endgroup$ – Mobassir Hossen May 23 at 15:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.