The performance or efficiency of a classifier is shown by various features that tells how well working the particular classifier is.
There are the various ways to check the performance of our machine learning model and why to use one in place of the other. We will discuss terms like:
Precision-Recall or PR curve
ROC (Receiver Operating Characteristics) curve
PR vs ROC curve.
Below some important performance measures which are used in machine learning:
This is also the same as the error matrix, by confusion matrix, it is easily shown that what percent of predictions made by our classifier was correct and where it was difficult for the classifier to predict the actual classification.
TP = True positive
TN = True negative
FP = False positive
FN = False negative
Accuracy = (TP + TN) / N, where N is sum of TP, TN, FN, FP.
This is the overall efficiency of the model
Sensitivity can be defined as the effectiveness of classifiers to identify positive labels. This is also known as recall.
Sensitivity = (TP)/ (TP+FN)
This is defined as the effectiveness of classifier to correctly identify negative labels.
Specificity = (TN) / (FP + TN)
Prevalence = (TP + FN) / N
N is the sum of all conditions i.e. TP, FN, FP, TN.
Positive predicted values
Positive_predicted_value = (Sensitivity * Prevalence) / ( (Sensitivity * prevalence) + (1 — specificity) * (1 — prevalence) )
Negative predicted values
Negative_predicted_values = Specificity *(1 — prevalence) / (((1- sensitivity)*prevalence) + (specificity * (1 — prevalence)))
DR = TP / N
Expected_accuracy = ( (TP + FN) * (TP+FP) + (FP+TN) * (FN+TN) ) / N
Kappa = (Observed accuracy — expected_accuracy) / (1 — expected_accuracy)
These are top and most important performance matrix which is used by every developer when predicting any machine learning algorithms.
Ref - https://medium.com/