Xente Fraud Detection Challenge
Helping Uganda
$4 500 USD
Completed (over 6 years ago)
Classification
2031 joined
545 active
Start
May 20, 19
Close
Sep 22, 19
Reveal
Sep 23, 19
score of 0 % repeatedly using different models on submission
Data · 12 Sep 2019, 11:52 · 8
I have gotten a score of 0 % on all the models i have submitted inclusing the xgboost, randomforest, svm, adaboost e.t.c. I would love to get a solution to this problem or anyone who also went through this problem and solved it should please help. Thanks
You must be submitting something wrong, for example, predict_proba instead of predict?
No, I used .predict() for all my models
Using Ensemble model can make it float data type convert to Integers it will work. 0.0 is not allowed use 0
I typecast the FraudResult column to int using the .astype(int) method several times. And I still got a 0% score
I got the same problem. Have you got a solution?
Hi there! I also once faced the same problem in this competition.
F1 score is the harmonic mean of precision and recall. And so, there is a tradeoff - as precision increases, recall decreases and vice-versa. The score of 0 probably means that your model(s) predicted only one class for all samples in the test data.
The solution would be to add a cost of penalization to the class of lower frequency, which is fraud(1). Models such as XGBoost, LightGBM, Random Forest and Support Vector Machines helps in such penalization.
Also, trying resampling techniques like SMOTE, NearMiss, Random Sampling, etc may also help.
Goodluck!
Hi there! I have been having a similar problem. Have you solved it yet?
be sure you are not splitting the data, rather encode them to avoid 0 score