Anomalies

1. Anomaly detection and visualization

Jupyter notebook

2. Irregularity (card fraud) detection

Supervised learning model applied: random forest

Parameter optimization: random search

Jupyter notebook

Strength: High accuracy

Weakness: Class-imbalance problem exists (only 0.172% of fraud labelled data)

Semi-Supervised learning model applied: One-Class SVM

Strength: learning is enabled without fraud labelled data

Weakness: Lower accuracy rate compare to supervised learning models

Autoencoder

Strength: No need of labelling pocedures

Weakensss: Low accuracy. Sensitive to hyper paramter