Credit Card Fraud Detection

Client Model

Machine learning model to detect fraudulent credit card transactions trained to minimise false positives and consequent loss of transaction fees

Client Model Information

Model Architecture (number of neurons)[784, 100, 100, 10]
Model Accuracy93.09%
Number of Parameters89,610

DC-COMP – Model Compression

  • Client uploads their model
  • Client selects one of the compression techniques from the list
  • Client sets parameters for the chosen techniques and model training processes
  • The compression technique is applied and the model is retrained
  • DC-COMP outputs the results table and graphs

DC-COMP – Results Table

Client chooses the compressed model based on criteria from the results table

DC-COMP – Result Graph

Model 4 has been chosen by the client based on their accuracy, size and speed requirements.