💳 Did my model catch the fraud? (Yes, 28 out of 31 times!)

I am excited to share that I have successfully built and deployed a Credit Card Fraud Detection System using Deep Learning.

In the world of finance, "Accuracy" can be a trap. If a model simply predicts every transaction as 'Safe', it might get 99.9% accuracy, but it fails its main purpose: catching the fraud.

💡 My Approach: I focused entirely on Recall (Sensitivity) rather than just accuracy. I wanted a model that minimizes False Negatives (missing a fraud case), because missing a fraud is far more costly for a bank than a false alarm.

🛠️ Technical Breakdown:

Model: Built a Deep Learning model using Artificial Neural Networks (ANN) with TensorFlow & Keras.

Imbalance Handling: The dataset was highly imbalanced, so I used SMOTE (Synthetic Minority Over-sampling Technique) to train the model effectively.

Preprocessing: Used StandardScaler to normalize transaction amounts.

Deployment: Created an interactive web app using Streamlit.

📊 The Results (Snapshot Attached): Testing on unseen data, the model performed exceptionally well:

Total Fraud Cases: 31

Detected Successfully: 28 ✅

Missed: Only 3

Recall Score: 90%

This project deepened my understanding of how to handle real-world imbalanced datasets and the power of Neural Networks in classification tasks.

Check out the live demo and code below! 👇

🔗 Live App: [https://creditcardfrauddetection-hgjvcczvktnrdpr3xoq5kt.streamlit.app/] 💻 GitHub Repo: [https://github.com/imukeshkumarprajapat/credit_card_fraud_detection]