Bram Cappers
Name: dr. ir. B.C.M. (Bram) Cappers
Position: Co-Founder AnalyzeData
Visualization Architect
Address: Eindhoven University of Technology. Department of Mathematics and Computer Science. Visualization group. De Rondom 70, Room MF 4.060, 5600 MB Eindhoven, The Netherlands
Phone: +31 (0)40 247 8863
E-mail: [email protected]
Looking for an interesting projects? Join our Arsenal @ AnalyzeData
Nowadays, every phone call is transferred over the Internet (Voice Over IP). It’s fast, cheap, and you don’t need a physical phone to get in touch with someone. Sounds great, until you receive a bill of ten thousand euros for a call you did not make... Telco fraud is a big issue and annually is responsible for millions of euros of damage. The idea is simple:

  1. you buy an off-shore premium number for which people have to pay e.g., 10 euros per minute;
  2. you find a phone device on the web with a bad username and password;
  3. you use the phone to dial your expensive number indefinitely.
At AnalyzeData it is our job to prevent as many fraud from happening as possible and you can help us!

We are a fresh startup located at the TU/e campus working on the next generation data analysis platform. If you are looking for the startup experience while working on something that really matters to the world, then you have come to right place :) Check out some of our projects. Do you have your own idea how to solve this problem? Feel free to contact us.
Assignments (suggestions)
Not on my Watch: Building your own online fraud detector

In order to cope with high data volumes in telco industry (in order of tens to hundreds multidimensional data points per second), speed and accuracy of fraud detectors is crucial. Our data analysis platform is designed to apply multiple machine learning detection techniques in parallel and classify data points in a highly-multi threaded fashion. This raises the question how to adjust existing anomaly detection techniques to work in such environments as efficient as possible. Have you always wanted to apply machine learning for yourself on real-world data? And do you think you can beat our system? In that case challenge accepted!

Discovering the unknown: Anti- VoIP fraud detection using Bayesian learning

In this project the student will investigate whether Bayesian learning can be beneficial for the early-stage detection of fraudulent calls. Given a real-world dataset the student investigates how to optimize different parameter settings for the chosen algorithm, which data features to include to improve detection rates, and the main limitations of the technique for this application. The main challenge: try to catch as many fraudulent calls as possible without making too many false accusations.

Identification of collective user activity: Clustering phone behavior

In the telco world there exists different types of phone users. On one end we have the traditional households calling once or twice per day to friends and family. On the other end it can be a callcenter dialing 10 calls per second simultaneously. Depending on this “user class” other detection techniques may be more fruitful than others. In this assignment the students investigate how we can identify these groups by clustering users with similar call behavior together. Given a real-world dataset the students investigate different type of distance metrics on how to cluster phone call records and investigate the relevance of certain features towards the clustering results. The main challenge: once we have obtained such clusters, can we predict the type of a user based on calls that we have not yet seen before?

Visual Analytics of Software container visualization

Modern cloud architectures virtualize their software in “containers” so that it can run regardless of the underlying hardware. Depending on the amount of data to analyze, more computing power may be required and additional machines need to be “spun up” dynamically to cope with data volumes. In this assignment the student becomes familiar with the Kubernetes container orchestration software platform and designs a visual analytics application that helps to better monitor (i.e., CPU/Memory) usage of running software containers. The challenge: can you figure out how to optimize our cloud platform (e.g., by putting multiple software containers on the same machine, detect unused resources etc.)?

Representation Matters: Feature Weighting strategies for improved detection

The Naïve insertion of high dimensional data in machine learning algorithms does not work. In order for such algorithms to produce practical results the effect of adding/removing/representing a feature differently needs to be studied. In this assignment the student designs a systems where the relevance of (combination of) features can be (visually) studied.

About AnalyzeData

No modern company can afford to not do data. If you are not leveraging your data to gain benefits, your competitors are! Unfortunately, data analyzes can be hard. Large upfront investments in time, staff, and money. We want to change that! AnalyzeData provides a broad spectrum of services to help you bring added value to your products/services based on data:

  • Detect and prevent fraudulent user logins and activity
  • Voice Over IP Fraud detection and prevention
  • Discover, Check, and Monitor your business processes

Interested in what we do? contact us and join our team!