Around 0.9808 Public LB Score Solution for TalkingData AdTracking Fraud Detection Challenge
The best single model we have obtained during the competition was an LightGBM model Public LB score 0.9808138 and Private LB score 0.9816301. Our final submission was a median ensemble of 5 LightGBM models which features and train data are slightly changed each other. This submission scored 0.9812967 on Public LB (159th) and 0.9821770 on Private LB (105th).
During this competition, I refered to some kernel and discussion, especially below ones.
Thanks for sharing sophisticated codes and thoughtful ideas.
https://www.kaggle.com/pranav84/talkingdata-eda-to-model-evaluation-lb-0-9683
https://www.kaggle.com/cgundlach/lgbm-single-model-lb-9791/
https://www.kaggle.com/asraful70/notebook-version-of-talkingdata-lb-0-9786
https://www.kaggle.com/c/talkingdata-adtracking-fraud-detection/discussion/53378#306201