How would you evaluate a logistic regression model?
Question
What are the different methods ?
in progress
0
Machine Learning
4 years
4 Answers
1061 views
Contributor 0
Answers ( 4 )
A subsection of the question above. You have to demonstrate an understanding of what the typical goals of a logistic regression are (classification, prediction, etc.) and bring up a few examples and use cases.
Evaluation of the model varies depending on your objective. It can be an accuracy score (for balanced data), f1_score, auc_score (for imbalanced data). In general, we should try to train our model based on log loss (cross-entropy).
there are various methods in evaluate logistic regression :-
– log-loss
– accuracy not preferred if data is imbalanced
– roc auc score
– precision or recall or f1 score
1) The most basic is the accuracy score which will give you a measure of how many records
were correctly classified, irrespective of the class they belonged to.
2) Another is AUC score, calculated from the ROC Curve which is a more complete representation
than the accuracy as it gives the trade -off between the true positives(positive data points that are
correctly considered as positive ) and false positives(negative data points that are mistakenly
considered as positive) at various cut points of probabilities. Depending upon the business problem
you are solving, you can choose an appropriate cut point.The area under this ROC curve is called
AUC score which ranges from 0.5 to 1. The closer it is to 1, the better is the classifier.
3) Precision is used when there is a high cost associated with False positives.
For example, Email Spam Detection, You do not want to miss an email which is
classified as spam but is actually non-spam.
4) Recall is used when there is a high cost associated with False Negatives.
For example, in credit default detection, if a person ‘likely to default’ is predicted
as person ‘not likely to default’ , it can create a problem for the lender.
5) F1 score is used when you need a balance between precision and recall and
you have a highly imbalanced data set
There are many ways to evaluate a model like Accuracy, Precision, Recall, f1 score roc , auc