Who are you?
Integrated Dashboard using Machine Learning and Image Recognition Technology
1) Difficulty in differentiating between legitimate and suspicious transactions
2) The databases are scattered and often stand alone or reside with different agencies
3) Suspicious Transaction Reports are reactive, not predictive or real-time
4) FIU do not have complete profile of the suspicious individual
An integrated dashboard that uses Machine Learning and Image Recognition technologies to flag suspicious person based on a set of indicators matched against past reported transactions to provide holistic view on the profile of the identified person(s).
With enough information and training data, the Machine Learning model will be able to learn and improve its indicators to conduct real-time analysis and identify suspicious transactions and person(s).
In this codeathon, there are several key actions we had identified at the start to create our solution.
1) Identify the flagging indicators, current modus operandi, and relevant data set
2) Understand relevant datasets to generate dummy data for the purpose of building our Machine Learning model.
3) Conduct preprocessing and feature engineering on our identified dataset by cross-checking and matching with other relevant data sets.
4) Build and test our machine learning model using our newly engineered data set.
5) Extract prediction results on transactions in order to identify suspicious individuals
6) Cross check the individual details and information against available data sources including FIU data bases, government data bases, web, and open sources.
7) Consolidate and visualize all these information into our 360 view dashboard along with our facial recognition technology
We also created an analysis and coding environment using JupyterHub for our team at http://220.127.116.11/. Our data preprocessing, dummy data generation and machine learning models are being generated there.
Our dashboard was developed using Ionic 3 Framework, leveraging on its Progressive Web App capabilities
You can check it out using the username 'knazran'. (No password needed).