This is probably the most widely used classification algorithm, and the first algorithm that a machine learning practitioner usually tries when given a classification problem. It performs well when data is linearly separable or approximately linearly separable. Even if it is not linearly separable, it might be possible to convert the linearly non-separable features into separable ones and apply logistic regression afterward.

In the following instance, data in the original space is not linearly separable, but they become separable in a transformed space created from the interaction of two features:

Also, logistic regression is extremely scalable to large datasets with SGD optimization, which makes it efficient in solving big data problems. Plus, it makes online learning feasible. Although logistic regression is a low-bias, high-variance algorithm, we overcome the potential overfitting by adding L1, L2, or a mix of two regularizations.