Other Key Industries
AI-powered analysis of millions of database queries
How can unusual patterns be identified and reliably evaluated in millions of database queries? For an international telecommunications company, HyPlus developed an AI-powered solution that models user behavior, analyzes anomalies, and automatically evaluates large volumes of data.
Summary
Industry: Telecommunications
Technologies: Machine Learning, Apache Spark, Big Data, Feature Engineering, SQL Analytics
Services: Data & AI Engineering
Customer benefits
- More than 1.5 billion data records and approximately 3 TB of data analyzed
- Machine Learning Model Developed to Handle Millions of SQL Queries Per Day
- Transparent Analysis of User Behavior Using a White-Box Approach
The Challenge
The company manages large volumes of sensitive data. With millions of database queries per day, it is becoming increasingly challenging to identify unusual activity manually or through rule-based systems. Therefore, the company sought a data-driven approach capable of automatically analyzing complex usage patterns and evaluating anomalies. At the same time, the solution needed to remain transparent. Business units and IT managers require clarity on why certain activities are assessed differently from typical user behavior.
Our solution
To analyze the enormous volumes of data, HyPlus developed an analytical approach that combines data engineering, feature engineering, and machine learning. To this end, SQL queries and supplementary metadata were analyzed, structured, and converted into analyzable features. This formed the basis for a machine-learning model that describes typical usage patterns of individual user profiles and evaluates new activities in terms of how they deviate from those patterns. Unlike rule-based methods, the developed approach is based on actual usage behavior and can therefore also evaluate previously unknown patterns and deviations. A central component of the project was the traceability of the evaluations. Instead of a black-box solution, a white-box-oriented approach was pursued, which makes it transparent which features contribute to an evaluation and why certain activities deviate from established usage patterns. To process the data, HyPlus developed a scalable architecture based on distributed data processing with Apache Spark.
The result
The solution automatically analyzes millions of SQL queries and identifies anomalies based on deviations from typical user behavior. Particular attention was paid to the traceability of the results. Business units and IT managers can see which characteristics led to a particular assessment and why an activity was classified as unusual. The result is a system that highlights suspicious database activity at an early stage.
Discover our digital solutions
You too can benefit from our extensive experience with projects involving companies of all sizes.
HyExpert
Would you like to know how our customized IT solutions can help your business succeed? Do you have questions about specific digital challenges?
Get in touch with us—we’re here to help!