I am training and testing the following five models for fraud prediction on a credit card fraud data set (available on Kaggle)

• Logistic regression (regular)
• Artificial Neural Network (regular)
• Cost-sensitive Artificial Neural Network
• Cost-classification Logistic regression
• Cost-classification Artificial Neural Network

All models are evaluated with 5-fold cross-validation in terms of both, F1-score and cost savings

main.py: Train, test and evaluate all models

eval_results.py: Function to evaluate results

ANN.py: Artificial Neural Network with custom loss function (built in Keras)

results.ipynb: Generate plots to visualize results

results folder that contains results generated by running main.py in .npy file format