Fraud Detection Report

In this project, we developed a comprehensive fraud detection model using a methodical approach outlined in five primary stages: data preparation, feature engineering, feature selection, model evaluation and implementation.

Workflow

Model Methodology:

  • Baseline Logistic Model
  • Decision Tree
  • Random Forest
  • LBGM
  • Catboost
  • Neural Network
  • XGB

Model Exploration

Final Model

The culmination of this project in the deployment of an XGBoost-based fraud detection model represents a significant step forward in safeguarding our operations against fraudulent activities. The model’s performance, evidenced by substantial potential savings and a strategic 5% cutoff point, highlights its effectiveness in identifying high-risk transactions with precision.

Download PDF file.