The client faced millions of dollars in losses due to fraudulent claims and needed to identify the likelihood of fraudulent claims, minimize false positives, and reduce manual review efforts.

Implemented an advanced predictive analytics approach to detect fraud.

1. Data Preparation: Received fraud labels from the client based on historical data and handled imbalanced data using SMOTE (Synthetic Minority Over-sampling Technique).

2. Feature Engineering: Extracted features from policy data (type, premium), customer data (age, income, vehicle details), and claims data (history, injury type, amount). Applied advanced feature selection techniques (RFECV - Recursive Feature Elimination).

3. Model Development: Built initial models using Logistic Regression (accuracy 72%, precision rate 63%). Enhanced performance with Decision Tree, Random Forest, KNN, and XGBoost, selecting XGBoost as the final model (accuracy 85%, precision rate 77%, false positive ratio 1:5).

4. Implementation and Monitoring: Automated model monitoring using customized Python packages, created Tableau-based dashboards for business stakeholders, and deployed the model on a cloud platform.

Reduced annual fraud loss by 22% and captured 15% more fraudulent claims. The advanced analytical methods and machine learning models implemented by Solytics:

- reduced annual fraud loss by 22%

- captured 15% more fraudulent claims, and

- automated the monitoring process.