search

roh2020 / Machine-learning-projects

solved credit card fraud detection probelem using only 7 features through machine learning and reached to an accuracy of almost 100% using stacking
2 stars 1 forks source link

Feature Selection is not working #1

Open akhileshkumargangwar opened 6 years ago

akhileshkumargangwar commented 6 years ago

HI, In your notebook how can you say that feature selection has worked. Data set is highly imbalanced and if it will miss classify fraud data, even though you will get 99% accuracy becauce fraud data is 0.08% (around https://www.kaggle.com/mlg-ulb/creditcardfraud) . Apply these claasification algorithm without feature selection and check how much accuracy you will get.
But you feature selection concept and insights seems good.

    Thanks
roh2020 commented 5 years ago

I have used stacking in my model in which i have combined the outputs of all the classification algorithms i have used and then i have done majority voting between them. Each classification algorithm i have used is giving accuracy of about 99% ,so when i combined all the algorithms and i can be pretty much sure that prediction will be error free.Moreover i have tested on some fraud data points and it is working well. Thankyou