Credit Card Fraud Analysis and Detection - - Anonymized credit card transactions labeled as fraudulent or genuine. The project was intended to detect fraudulent transactions from a highly imbalanced dataset. The project was intended to detect fraudulent transactions from a highly imbalanced dataset. Performed data analysis and developed machine learning models for a credit card fraud detection using ULB credit card dataset.
- To solve the imbalance dataset problem random undersampling, oversampling and SMOTE techniques were used.
- Performed data cleansing (with the help of correlation matrix, box plot and interquartile range) and dimensionality reduction using PCA, t-SNE and truncated SVD.
- Created Logistic regression (0.84), support vector machine (0.90) and decision tree (0.94) and neural network (0.97) based classifier. F1scores along with ROC cure was used to measure the performance generalization of various classification models.
- Neural network model with SMOTE based oversampling generated the best model with an F1 score of 0.97.
- Also created a Gaussian distribution based anomaly detection model with an F1 score of 0.65.
- Models -Logistic Regression -Support vector Machine -Decission Tree -Autoencoder