This kernel including how i did feature engineering and the model building parts. Since the competition is confidential on Kaggle, please contact me if you want the dataset.

2018 Travelers Statistical Modeling Competition: Claim Fraud Detection

Business problem: Imagine you work for our fraud detection team in Claim department as a modeler. Your task is to create a predictive model based on historical claim data. Your team is interested in the key drivers of fraudulence and wants to achieve an accuracy as high as possible. For this case competition, your group is tasked with identifying first party physical damage fraudulence and explaining reasons of fraudulent claims.

Modeling Rules:

• Individual and team participants are both allowed. If you will team up, you can either log in using your personal account and then use the TEAM function or sign up as a team.

• Each group can make at most 3 submissions per day.

• Work together within your group, but not between groups.

• You can build the model using any software, python/R/SAS are preferred.

• Work together within your group, but not between groups.

• Work with your group members and ask for clarification in the DISCUSSION section.

Competition Rules: Top 5 teams from each campus will give a presentation to explain their models and findings. The winning team will be invited to Travelers Campus Day.

The evaluation metric for this competition is AUC score. AUC score means Area Under Curve. The curve we used for this competition is ROC curve.

You can use 'pROC' package for r or sklearn.metrics.roc_auc_score for python to calculate AUC score for your model.