Dear Monika Sharma ,

It is quite good, your learning curve training set size is 400, training accuracy is 1.0. When your Training set size is increased the cross and validation score increases rapidly and end of the learning curve the training set accuracy and cross and validation accuracy are the same. The highest accuracy is achieved 1.0 which is really nice.

The first impression is that it works very well by giving 100% accurate results. However, if we look at the curve jumping to almost 98% for the first 60 samples and slowly jumps to 100% for the remaining 400 samples means that there is something wrong. Because the last 2% accuracy depended on about 90% of the samples. It means that at the beginning the algorithm converged fast to the local optimum solution that should not be the case

J. Rafiee

Md. Ashiqul Islam thank you

Mohamad M. Awad, Thanks for the explanation, I also suspected something a miss. When I used LogisticRegression with the same setup, I got a different Curve. (Attached the curve)

Now the question is, whether I should go with LogisticRegression and try to make it better as it is not converging completely, or shall I try to overcome the local optimum issue happening with Decision tree/ Random Forest? Please Guide

Thanks a lot

Yes, you can try the three algorithms and compare them. You can improve RF or any Trees-based algorithm by changing some parameters such as The number of decision trees, The maximum number of leaf nodes in each tree...etc.

But LogisticRegression is reasonable to me good luck Monika :)