endorsedCompleted

Anomaly Detection ML System for Acceptto

Services Covered:

Custom Software Development

Industry

Cloud & Network Security

Duration

10 months

Budget

Confidential

Start Date

1 September 2018

Team Size

1-5

Engagement Model

Offshore

Project Capability Score (PCS)

74/100

Good

Project Capability Score (PCS) estimates how capable Helpware Tech is of successfully delivering a project like this, based on its past experience, track record, and ability to handle similar work. Learn more about PCS

Client

Acceptto

Portland, United States of America

Project Summary

Helpware Tech (formerly eTeam) built a machine-learning system to analyze user behavior, detect anomalies, and cluster similar users for a cybersecurity client. The vendor's data scientist used unsupervised learning techniques implemented in Python with NumPy, Pandas, and Scikit-Learn and integrated those into a tested auxiliary package. To compensate for a lack of real data, the team developed an artificial dataset generator. The internal team reported strong communication and successful, dynamic deliverables.

Key Challenges

Create a machine learning system to handle anomalies in user behavior
Analyze user behavior to detect anomalies
Cluster similar users to control behavior changes
Lack of real data for training models

Project Deliverables

Machine learning anomaly detection system
Unsupervised outlier detection and clustering models
Scikit-Learn pipeline with integrated side packages
Auxiliary Python package with unit tests and type checks
Artificial dataset generator based on Dirichlet distribution

Project Solution

The vendor's data scientist performed an analysis of user behavior using unsupervised learning for outlier detection and clustering to identify user groups and behavior changes. Development was done in Python with NumPy, Pandas and Scikit-Learn; side packages were integrated into a Scikit-Learn pipeline and main scripts were combined into an auxiliary Python package with unit tests via pytest, type checking, and documentation. The vendor provided regular insights and progress updates throughout the engagement.

Project Outcome

Reached project goals despite lack of real data
Added a visualization module with metric selection
Compared algorithm efficiency and produced user clustering with cluster visualization
Developed an artificial dataset generator with controlled variance changes and feature grouping

Platforms

AI/ML Platform
Web

Tech Stack

Python
NumPy
Pandas
Scikit-Learn
Pytest

Client Endorsement

Overall Review Rating

Timeliness

Cost Rating

Willing to Refer

Quality of Deliverables

“The internal team found the deliverables to be dynamic and successful. Helpware Tech (formerly eTeam) used their creativity and problem-solving skills to create an artificial dataset generator to compensate for a lack of real data to rest the product. Strong developers, they handled communication without issue.”
Stefan Ulbrich
Principal Research Scientist
Date: 09/19/2019