Banking Case Study: Fraud Detection

Dashboard presents results of our model regressor which predicts whether a customer transaction is fraudlent in a banking sector.
Following ingestion from multiple sources, we explore multi-characteristics of the data, build our model and finally present the results.

Characteristics of the data exhibiiting multi-characteristic distributions

Category Distribution

Demographic Distribution

Age Distribution

Fraud / Non Fraud

Fraud

Non Fraud

Algorithm & Model Building

Feature Importance

Receiver Operating Characteristic

AUC - 0.94

Ingestion of the data from multiple data sources

Source 1: demogrpahic data

Source 2: expenses data

Source 3: fraud data

Model Results

Test Table

Probability Prediction on Few Test Samples

Non Fraud

Fraud

Management Recommendations

1. Avoiding fraud transcations is the number one business goal.
2. Banks need to predict which transactions are fraudlent to avoid costs.
3. We recommend the bank to consider Features Importance chart which are strong indicators to pre-empt fraud.