This is an old revision of the document!


Master Projects

Documents

Possible assignments

Data Science: Developing a Self-Standing Dynamic reporting tool

A huge amount of (transaction) data is generated on a daily basis in ASML Development and Engineering department. The data is scattered in different sources. The challenge would be extracting data from relevant sources and creating a self-standing dynamic reporting tool (dashboard) demonstrating performance of (Supplier Quality) Engineers in different granularity levels (Department, Section, Individual) based on a set of pre-defined KPIs.

Are you a master student in Software Engineering (Data Science) with a passion on real-life data challenges? Then we are looking for you!

→ Read more...

Real-Time Model Discovery of the Service Order Process Using Stream Process Mining

Kropman Installatietechniek is a Dutch company established in 1934 and has become one of the leading companies of the Dutch installation industry. With about 800 employees, 12 regional locations and an annual turnover of more than 100 million Euro, Kropman is an integral service provider with a multidisciplinary approach. Kropman is mainly active in office buildings, health care and industry. It offers design, construction and maintenance in the field of facility installations. Kropman also has a separate business for process installations and cleanrooms: Kropman Contamination Control. The maintenance (services) is a fast growing business line. The order process is fully supported within an ERP environment. The service order process is not a trouble-free process: the process takes too long and flows more often than necessary. The company aims at increasing the throughput of the SO process and decreasing the amount of process deviations by applying process mining and data mining techniques.

To have a better overview of the process, Kropman is aiming at:

  • Exposing bottlenecks happening in the SO process
  • Detecting deviations from the supposed model in the real time they are happening using stream process mining techniques

→ Read more...

Understanding Customer Journeys with Process Mining

In today’s customer environments, where customers use many different contact channels to solve outstanding questions and do requests, it becomes increasingly difficult to follow customer behavior, optimize service levels and provide a memorable experience for customers. Especially when the contacts of the customer are not related. These journeys can be product specific (e.g. changing your telephone provider), customer specific (e.g. change known home address) or life event specific (e.g. starting a family).

Underlined works together with several companies like CZ, Aegon and SNS to build a generic framework in which all traceable, incoming and outgoing customer contacts (call, web, e-mail, chat, etc.) are brought together as a unique dataset. The dataset of companies is further enriched by Underlined with relevant analysis that can be linked to (unique) customer events.

Underlined has worked together with the TU/e (Bart Hompes and Joos Buijs from the Architecture of Information Systems group) and CZ (one of the largest Dutch health insurance companies) to develop customer journey mining algorithms. This research showed that it is possible to distinguish the different journeys per customer without any prior process knowledge, but additional research is needed to:

  1. Apply machine learning techniques to cluster customers with similar behavior. The customer journey can significantly vary depending on the customer characteristics. Therefore, try to build a single model for all customers will lead to not fully satisfactory results. Therefore, the application of machine learning or OLAP techniques (a.k.a. as process cubes) would be beneficial to split the customers into clusters each containing customers with similar characteristics.
  2. Research multi-dimensional similarity matrix. Current customer journey mining algorithms need to know which activities might be related to each other. Information on this is stored in a similarity matrix. Currently there is a working version for relatively simple datasets and processes. We would like to develop a next level similarity matrix for more complex data which contain multiple journeys and multiple customer segments.
  3. Predictive modelling of emotions in the journey. Customers make decisions in the customer journey based on their emotions. In current datasets there is sample feedback, which expresses the feedback of customers and their concerns regarding the service of a company. We would to research a predictive model that, out the basis of the past customer journeys, it can predict the intermediate and final emotions of the customers who are currently active in their journeys.

In the above-mentioned analysis, the student should try to leverage on every piece of available data. This includes logged data of the past customer behavior, for example call center data, online click trails, social media data and online and offline feedback. But also non-transactional data like product usage and customer segmentation variables.

→ Read more...

Improving the Evolutionary Tree Miner

The Evolutionary Tree Miner (ETM) is a genetic process discovery algorithm that works on process trees, a specific process modelling formalism. Recently work has started on an interactive process discovery algorithm where the user is guided to modify a free-choice Petri net in such a way that the Petri net is always sound.

The master project would consist initially of ‘connecting’ the ETM to this interactive process discovery approach, hence replacing the human. The main benefit would be that the ETM does not solely work on process trees any more (which can be somewhat restrictive), but directly on Petri nets in such a way that they are guaranteed to remain sound.

After this initial step several other improvement steps can be applied to the ETM such as extending the work in deriving alignments, estimating quality of a process model, smarter mutations, smarter application of mutations, starting from solutions created by other algorithms, etc. etc.

Therefore an important part is the implementation of these ideas in the ETM which is implemented in our toolset ProM. Good programming skills in Java are therefore important, no matter if you are a BIS or a CSE student.

→ Read more...

SAP Process Mining at Ciber

Ciber Netherlands [1] is an IT consulting company, with its origins in Detroit, integrated now within the Manpower group. They have a strong interest in using Process Mining to improve their systems and the services they provide to their customers. In particular, they wish to apply Process Mining to the systems they use for managing their internal processes. Their interest is on focusing their efforts on the SAP platform, which many of their clients use as well. Performing this project on their own SAP systems would be a great way to demonstrate the potential of Process Mining in these kind of environments, which they would be able to extend for their clients. Until now, Process Mining on SAP systems has been performed in an ad-hoc fashion. However, we aim at automating this procedure and applying new techniques [2,3] to extract the relevant information. These techniques allow to retrieve historical and execution information, enabling the application of analysis methods in a more standardized and meaningful way. To apply these techniques in real-life scenarios, many challenges remain open. One of them is to be able to identify interesting views on the data in order to obtain useful representations of the process. Often, this requires the involvement of domain experts, but a more automatic way would be desirable. Therefore, the project would not only involve the application of existing techniques, but the development of new methods to identify meaningful views on the real data.

→ Read more...

Océ is a Netherlands-based company that develops, manufactures and sells printing and copying hardware and related software. Their cut-sheet production printers are used to produce millions of prints on a daily basis. Print jobs may involve printing on different paper stock, which is loaded in the various paper trays of the machine. Print operators are mainly busy with loading the printer with new paper and unloading the printed paper stacks. As this is a human task, errors are made in the process, especially in loading the paper trays with the correct paper stock.

Using the wrong media for a print job may lead to print quality issues and is most often unacceptable for the print buyer. For these reasons, a timely and automatic detection that the wrong media is loaded is needed to guarantee the overall production quality.

Your assignment is to use machine learning to determine whether the loaded paper media in a paper tray actually matches the medium specified on the printer. As input, a set of data can be used that is logged during the printer operation.

→ Read more...

Older entries >>

See also the overview of completed master projects.


If you have a master project item for this page, which may include possible master project assignments, on-going master projects, and completed master projects, please send it to Eric Verbeek.