Using financial dataset with real anonymized transactions to highlight suspicious activity using machine learning.
I was able to predict suspicious behavior with a near perfect accuracy score. Due to the near perfect accuracy scores produced by the models, I worked to see if there was overfitting in the models by implementing the ROC curve the AUC. In addition, I tested to see if the imbalance in the prediction feature is affecting the results. The imbalance is due to our prediction class being very small as our suspicious poplulation is mimicking the suspicious population in reality.