The latest articles on Forem by Retro (@retro099). https://forem.com/retro099 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3810581%2F9f4a3da1-27ed-4628-9276-4d3a6e32dfe5.jpg https://forem.com/retro099 en Retro Fri, 13 Mar 2026 15:44:25 +0000 https://forem.com/retro099/i-dockerized-a-bank-grade-credit-card-fraud-detection-app-with-xgboost-recall-092-shap-320o https://forem.com/retro099/i-dockerized-a-bank-grade-credit-card-fraud-detection-app-with-xgboost-recall-092-shap-320o <h1> I Dockerized a Bank-Grade Credit Card Fraud Detection App with XGBoost </h1> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fmeexqv9eaenmepoby0ye.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fmeexqv9eaenmepoby0ye.png" alt="SHAP Force Plot - Fraud Cases" width="800" height="178"></a></p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fc0fj7s6y8633d8hhmvy0.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fc0fj7s6y8633d8hhmvy0.png" alt="SHAP Summary Plot" width="800" height="994"></a> </p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Forq2vqoscc6ytz2xmptk.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Forq2vqoscc6ytz2xmptk.png" alt="Confusion Matrix - Best Model" width="800" height="633"></a></p> <h2> 1. Project Background &amp; Challenge </h2> <p>Credit card fraud detection is a classic <strong>extreme imbalanced data</strong> problem (fraud rate only 0.172%).<br><br> Normal accuracy looks amazing (~99.8%), but in real business the cost of <strong>false negatives</strong> is huge.</p> <p>So I built the model with a <strong>Recall-first</strong> approach.</p> <h2> 2. Key Results </h2> <ul> <li>Recall: <strong>0.92</strong> </li> <li>PR-AUC: <strong>0.85</strong> </li> <li>SHAP analysis clearly identified <strong>V14 and V17</strong> as the top fraud drivers</li> </ul> <h2> 3. Tech Stack &amp; Production Features </h2> <ul> <li>Model: XGBoost + scale_pos_weight for imbalance</li> <li>Production: Docker + docker-compose</li> <li>Testing: Full unit tests</li> <li>Model persistence: joblib with DataFrame input</li> <li>Dependencies: All strictly pinned</li> </ul> <h2> 4. How to Run (Docker) </h2> <div class="highlight js-code-highlight"> <pre class="highlight shell"><code>docker compose up </code></pre> </div> <h2> 5. GitHub Repository </h2> <p><a href="https://github.com/Retro099/ML-Projects/tree/main/Credit_Card_Fraud_Detection" rel="noopener noreferrer">https://github.com/Retro099/ML-Projects/tree/main/Credit_Card_Fraud_Detection</a></p> <h2> 6. What I Learned </h2> <ul> <li>Always use Recall + PR-AUC as main metrics for imbalanced data</li> <li>In production, always use DataFrame for predictions</li> <li>Dockerization dramatically increases portfolio credibility</li> </ul> <p>If you found this project useful, feel free to star the repository ⭐</p> showdev machinelearning docker xgboost