What it does

A Machine Learning model, leveraging banking transactions data to detect fraudulent activities. The system first detects the potential fraudulent bank account and then it provides data visualization tool to manually investigate whether the bank account labelled as suspicious is really suspicious (i.e. whether it is true positive). This is quite important since the costs to check a suspicious bank account are very high, so it is not worth spending a lot of money on false positive examples.

How we built it

The predictive model is based on Regression Trees and Neural Nets. We used the modern libraries as Keras and Skikit learn. For the visualization part, we created a simple Flask server which loads and plots the data using the Javascript libraries d3.js and plotly.js.

Challenges we ran into

To make effective data preprocessing and data exploration and then to find the most suitable hyperparameters.

Share this project:
×

Updates