← All work
INTERNSHIP · MLCodSoft internship project
Credit Card Fraud Detection
A supervised classifier for detecting fraudulent credit-card transactions, built during the CodSoft internship.
Role · ML — classification on imbalanced data
PythonPandasScikit-learnSMOTE
The problem
Fraud detection is an imbalanced-classification problem: genuine transactions vastly outnumber fraudulent ones, so a naive model can look accurate while catching almost no fraud. The work here is about handling that imbalance honestly.
Approach
A standard, careful classification pipeline.
- Exploratory analysis and preprocessing of the transaction data in Pandas.
- SMOTE oversampling to rebalance the training set.
- Model training and evaluation with Scikit-learn.
Work in progress
A full write-up with the evaluation methodology is in progress; no performance numbers are quoted until they're confirmed.