It is a api or a library for python for implementing machine learning algorithms by directly declaring the classifiers and training the data on them.By doing so you can generate a model and then just use that model to predict the values.Scikit learn is opensource library and is contributed by many developers and because of which it has best algorithms which are implemented.Almost all the algorithms can be easily used by single line of code and also the parameters can be modified according to your requirement so it is the best library.

Scikit-learn is the most wanted library for python for any machine learning engineer for machine learning project.If you are experienced in ML and have little less knowledge about its implementation then you can use this because here you can create any classifier or regression model just by calling its object.This object can be trained by your training set and this ready trained model can be used to predict the furthur results.The other benefit of it is that if you want to change the parameters of the particular algorithm than also it can be changed by calling the object and passing the necessary values.It also has very clean documentation which is very easy to understand.

You can do classification, clustering, regression, pre processing and so many. If you are working in machine learning based research, I would highly recommend this package.