Need to fill the gap in your tech team?

Rapidly become your extended team or build a product from scratch. Top-notch engineering solutions by Uinno.

learn more

Anomaly detection with ML and deep neural networksML solution for world-famous startup

ml

offeredservices

Machine <br /> Learning
Machine
Learning
Predictive <br /> Analytics
Predictive
Analytics
Behavioral <br /> Analysis
Behavioral
Analysis
Deep Neural <br /> Networks
Deep Neural
Networks

techstack

Sklearn
Sklearn
Keras
Keras
TensorFlow
TensorFlow
Python
Python
Amazon S3
Amazon S3

clientoverview

overview

The Project

Our client is an extremely popular worldwide social media platform where influencers can share exclusive content with their fans through regular subscriptions and direct paid private messages. Fans, in turn, can send extra tips as an act of deep appreciation. The core value is the opportunity to form direct communication between celebrities and their followers that is beneficial for both parties.

How We've Accepted Non-Trivial Challenge

There are millions of active users and the platform continues booming. But as usual, popularity attracts people with bad intentions who are willing to earn by-passing the rules. Over time there were more and more cases that got noticed when a model and a fan collude. The follower makes donations and the creator withdraws money. After a while, the fan claims it was an unauthorized payment due to a stolen card and makes a chargeback at the platform's expense. It was impossible to distinguish such fraudulent users out of available analytics so the company's management decided to implement preventive measures.

designingoriginal solutions

core_solutions

The Uinno team was engaged in the development of a specific part of the social media platform. We were challenged to beat the fraud pattern and to not only seek for violators among the existing userbase but detect them before they even start to pay.

1. Analyzing the fraud pattern

1. Analyzing the fraud pattern

Based on the existing data of fraudsters and ordinary users, we have utilized deep neural networks technology to define certain patterns of activities performed by fraud users. The determined chain of actions got gathered into a pool of fraud behavior. Using machine learning, we are able to prepare a certain model for learning, input the gathered fraud pattern, and further analyze the predicted user behavior. It is impossible to so with standard analytical tools.

2. Developing neural network models

2. Developing neural network models

The central part of this whole process is neural network model development. We create various algorithms for neural networks based on TensorFlow, SK Learn, Keras, and other narrowly focused ML frameworks. A combination of those frameworks allowed us to process huge amounts of data available in this social media, to learn under high loads, and analyze information. In general, we've used a variety of neural network architectures like recurrent and convolutional neural networks. Then we've launched each developed model, tested and fine-tuned its parameters to make it ready for further data predictions. The process of these models’ development is continuous and relies on up-to-date data.

3. Implementing predictive analytics

3. Implementing predictive analytics

The developed by Uinno ensemble of neural network models allows allocating fraud users based on the defined pool of fraud actions within a few days after registration. The system gathers certain characteristics from the behavior of new users, defines the most appropriate analysis algorithm from the existing ones, stores it into a specific bucket list with Amazon S3. Then the information is processed using the developed neural network models. Each one is involved in a different stage of data analysis supplementing each other. This way, the system can detect which user is indeed a fraud or not. Once there are some specific actions noticed, the account may be limited or even banned.

Collaboration model

We have fine-tuned the collaboration with the client over the negotiated roadmap for a certain period of time that contains all the necessary tasks we have to deliver. Occasional checkpoints allow us to sync up with the client to show that we are working within the defined plan. The scope of work may be adjusted according to the updated information or the need.

What's in theoutcome?

outcome

Uinno has developed a comprehensive fraud detection solution based on a custom user behavior scoring system combined with traditional machine learning methods and deep neural networks for high-load platforms including:

  • The business logic of user behavior analysis;
  • The data model of behavioral users activity;
  • The pipeline of predictive model development.

The solution developed by Uinno allows revealing more than 80% of fraudsters even before users commit fraud that has extremely reduced the number of fraudulent cases.

Taking into account the amount of new users increment which is nearly 200,000 - 250,000 per day, our effort has saved much time and money for the client compared to manual processing of anti-fraud actions. Not to mention, the number of saved funds that could have been stolen by the fraud pattern.

The provided solution may not only serve the anti-fraud proceedings but help to establish better marketing efforts via a list of personalized influencers’ recommendations and overall improvement of related company services.

explorenext project

arrow-right