Company X is a company that runs a web based marketplace for events. They create a platform that makes it easy for event hosts to list their event, sell/distribute tickets, and collect revenue. Company X is experiencing problems with fraudulent event listings in their platform. There have been cases where users have bought tickets to fake events and payouts have been made to fraudulent hosts. In these cases, Company X is responsible to reimburse their users for lost value (alternatively they can choose to lose public trust and shut their doors). To avoid this problem, the company is now dedicating costly resources to investigate potential fraud listings and shut them down.

As such, the task is to provide a scalable solution to handle fraudulent event listings.

• Explore available data
• Perform EDA to determine signal in various features
• Develop pipeline to encapsulate cleaning, transforming, and splitting of data
• Train a suite of models and compare accuracy
• Select top performing model and tune hyperparameters to optimize for recall
• Create a dashboard show the results.

• Develop web site to present visualizations on the problem (link: http://54.84.68.72:2222/)
• Include API in the website to receive new live data from client
• Create database to store data
• Call data from database, clean it, and run model to generate a prediction and present it on the web site

• Number of Observations: 14337

• Number of Variables: 43

• Fraud Rate: 9%

• Variables used: 10

• Engineered Variables: 21

Chose 5 different models for this project:

• Random Forest
• Gradient Boosting
• AdaBoost
• Bagging
• Decision Tree

Gradient Boosting performed the best so we will focus on this model.

• Accuracy : 0.956
• AUC Score: 0.952
• AUPRC : 0.807

Precision and Recall vary based on the chosen threshold therefore they are not listed.

Here are is the Feature Importance Graph (Some variables can be removed since we can clearly see they did not provide any value to our Model)

The prediction probabilities were analyzed to determine valuable thresholds for high, medium, and low risk.

• High risk events can have an automatic pause put on their account until an expert at the company can review it more closely.
• Medium risk will be flagged and events will be reviewed by the investigation team at the company.
• Low Risk events will be be ignored.

• High Risk >= 0.6
• Medium Risk >= 0.062
• low Risk < 0.062

• Captures 52.2% of all fradulent Activities
• Requires 5.1% of all transactions be looked at
• Out of that 5.1%, 91.3% will be fraud

• Captures 86.9% of all fradulent Activities
• Requires 16.2% of all transactions be looked at
• Out of that 16.2%, 47.6% will be fraud

Created a Dashboard using Tabluau that allows the client to chose their desired threshold and average fraud cost

Access the dashboard here