Detailed Plan
What is this about?
Building a Machine learning Model to manage risk on Volibra -Decentralised Payment System.
One of the most critical requirements for operating a healthy and trustworthy decentralised payment system is ensuring the confidence of the merchants in the payment network, as well as their customers.
Investigating abnormal transaction activity requires a more complex analysis and significantly more time. Traditional financial systems and regulators are constantly struggling to identify anomalies and separate true from false positives. Today's DeFI players are not left out from these challenges. What happens when Crypto payment becomes mainstream?
The possibility of automating this process by extracting specific transaction activity could provide DeFI teams on Cardano with the opportunity to react faster to incidents and to identify abnormal patterns in a more precise manner.
The goal of this project is to build a model that can identify this risk in DeFi System.
The Plan: The Model ,Roadmap and Milestones
Approach: capture, analyse, visualise and Alert
Model-(Volibra Similarity Score)
A bit of context for CA/VCA with minimal knowledge with regard to Machine learning. Check Machine learning Clustering model on Google.
Volibra Similarity Score will be a machine learning clustering model that will identify patterns of disruptive or manipulative transaction behaviour.
Volibra Similarity Score will be a clustering algorithm that slices payment activity data into clusters based on time, merchant ID, transaction type and the time proximity of other payment actions etc.
As each cluster represents a time slice of payment transaction activity, the ML algorithm will classify this activity in a specific category that is a representative of the merchant's actions for that time period.
These transaction activity clusters may be thought of as packets of intent since each cluster contains a group of payment actions (payment request, refund ,modifying, or canceling) that may well be related, given the time proximity of these events.
Our clustering approach will offer a better view to investigators since the full context of the potential abnormal transaction behaviour will be captured, analysed and visualised.
Each Cluster model targets a specific category of abnormal or manipulative payment transaction activity.
The concept of the similarity score addresses this problem, by scoring each alert based on the degree of quantitative similarity to past actions. This similarity score is generated for each cluster on a scale of 0 to 100, pointing to the clusters that have the highest risk of drawing future regulatory attention and therefore are the most important for immediate review.
RoadMap
1. 1-3 months post funding
February 2022
- Define Requirement
- Set minimum acceptable performance metric
- Set testing criteria
March 2022
- Data collection and data engineering
April 2022
- Build Data Pipeline
2. 4-6 months
- complete sanity checks
- clean the data
- create new features for validation and test datasets
- Explore ML clustering Model
- build model
Q3 2022
- Debug, analyse, and tune
- hyperparameter tuning
- Validate your model
- Test Model
- Evaluate performance
Q4 2022
- Run public testnet experiments
- Integration with Volibra
- Build Analytics dashboard
- Monitor and maintain
Definition of Success
1. After 3 month:
- Testnet Data collection has been completed
- Data pipeline has been built
2. After 6 month:
- Analytics dashboard has been built
- Risk Model has been built and deployed
3. After 12 month:
- Different Risk pattern has been Identified
- Monitor, Maintain and improve the Risk Alerting
**Impact:**singularityNet and Cardano
The model will be publish in singularityNet marketplace platform for other team to test, this will help team to streamline their compliance and risk reviews internally.
Budget Breakdown
- 80 engineering hours - Risk Evaluation and ML Risk modelling on-chain transaction data - 8500
- 65 engineering hours for data pipeline and Data Engineering - 5500
- 110 engineering hour for Analytic Dashboard Development - 12500
- GPU Cluster 3500
Total: 30000USD
Lauch Date: Q4 2022
Team
Machine Learning Researcher - Alexandra Uma
-PHD Student - Machine Learning
Specialises on neural network models ,Neural Networks for NLP (NNNLP),Text Analytics
Classification Models in Economics
<https://www.linkedin.com/in/alexandra-uma-78255353/>
Data Scientist and 6+ Experience Python Developer - Dejene Techane
Specialty includes: Data Wrangling ,Classification Models,
Segmentation and Clustering,Predictive Analytics for Business
<https://graduation.udacity.com/confirm/GFUDQ2YW>
<https://www.credly.com/badges/aeee67f2-74e2-4788-94d9-365f907d95e7>
Lead Developer: Jude Ben - 9 years+ Software Development , Plutus Smart Contract Development , Blockchain and payment System, Cloud and Infrastructure Engineer
MSC thesis on Reinforcement Learning
Udacity Nanodegree : Deep Learning
<https://graduation.udacity.com/confirm/MSGHFEKN>
<https://www.linkedin.com/in/judeebene/>
Launch Date Q4 2022
