Credit-Card-Fraud-Detection

The dataset was imbalanced

Algos tried on both imbalanced and balanced dataset with close notice on precision,recall & f1-score

• Shallow Neural Network
• GradientBoostingClassifier
• SVM
• Logistic Regression
• RandomForest

comparison of the models (for fraud)

Models tried	P	R	f-1
shallow_nn on val	0.68	0.78	0.73
GradientBoostingClassifier on val	0.67	0.67	0.67
SVM on val(1000)	0.66	0.64	0.65
Logistic on val	0.73	0.53	0.61
Random forest on val	0.81	0.47	0.60
SVM on val(5000)	0.18	0.83	0.29

comparison of the models on balanced dataset

Models tried	P	R	f-1
shallow_nn on val
Not Fraud	0.89	1.00	0.94
Fraud	1.00	0.87	0.93
GradientBoostingClassifier on val
Not Fraud	0.94	0.92	0.93
Fraud	0.92	0.94	0.93
SVM on val(1000)
Not Fraud	0.96	0.93	0.94
Fraud	0.93	0.96	0.94
Logistic on val
Fraud	0.96	0.93	0.94
Not fraud	0.93	0.96	0.94
Random forest on val
Not Fraud	0.99	0.92	0.95
Fraud	0.98	0.91	0.95

Conclusion:

I feel SVM to be best suited!!