Card Fraud Strategy & Analytics

Card Fraud Strategy & Analytics

1. Applied image augmentation techniques, such as histogram matching and random biased crop, on brain scans to alleviate the problem of imbalanced data and lack of data; increased recall score of the final model by 0.052. Implemented a deep U-Net Convolutional Neural Network architecture with Focal Tversky Loss and RELU activation function using TensorFlow on GPU to perform lesion segmentation; achieved a recall score of 0.88

1. Analyzed 50M+ card transactions via SQL in Snowflake and identified the root cause behind the delay of hold release; collaborated with cross-functional team to reduce the clearing time from 2 hours to 49 seconds per case2. Designed metrics related to customer count and payment volume with customer registration and transaction data to evaluate campaign performance; built a dashboard via Tableau to track and visualize these metrics3. Built an ETL pipeline to combine card transaction data from multiple sources and performed data validation

1. Constructed a relational database consisting of 200K records of company transactions integrated from multiple data sources using MySQL; developed a predictive model using Neural Network in Python based on these records to identify prospective clients, resulting in a 15% increase in customer conversion rate2. Enhanced the current risk monitoring system by applying Random Forest and Gradient Boosting tree model to filter the most impactful credit risk features and implementing an automatic data anomaly detection program for these features; reduced the percentage of non-performing loans from 4.1% to 1.5%3. Developed a loan application review system to conduct qualification assessment for small business loans by using classification models, such as Decision Tree and XGBoost, reducing loan review time by 15%4. Applied a Tobit statistical model to predict borrowing demands of existing loan clients; dynamically adjusted credit limits with consideration of both estimated demands and risk profiles of clients; achieved a 12% increase in total loan amount while maintaining low credit risk