Philips HUE Product Evolution Using Stream Mining of Customer Journey

Philips HUE is a connected personal lighting system. It is controlled by a range of apps and smart home devices.

To acquire Philips HUE, one starts with a starter kit that consists of a few lamps and a bridge. Subsequently, consumers decide to expand their system with additional lamps or/and physical sensors. About 50 lamps can be controlled in one system.

An example of HUE system is seen in the Figure below. It consists of a bridge, three color lamps and a dimmer switch.

Every consumer chooses a device to interact with the lights. By default, all consumers start with HUE or third-party app. There are also options in the HUE app to create routines to be able to control the lights automatically. For instance, one can create a wakeup routine so that the lights in the bedroom go on naturally in the morning. HUE can also be controlled by Philips physical sensors (motion sensor, tap, dimmer switch), voice control and other third party smart home devices. Amazon Echo, Google Home, Eneco Toon, Homekit are some of smart home devices that are linked to HUE. In addition, traditional wall switch can also be used to control the lights. For further information see Philips Lighting website.

The master thesis focusses on developing a process mining model to understand product evolution of consumers in their journey and identify the paths that contribute to a high Customer Lifetime Value (CLV). Moreover, the thesis will also explore data mining techniques that can be applied on the outcome of process mining model.

We would like to capture customer behavior in process models. This way we can understand the behavior better and systematically explore ways of influencing this. Process mining techniques can be used to extract process-related information from event logs. Process mining is the linking pin between data science and process science. Data science approaches tend to be process agonistic whereas process science approaches tend to be model-driven without considering the “evidence” hidden in the data. Process mining aims to bridge this gap. Process mining seeks the confrontation between event data (i.e., observed behavior) and process models (hand-made or discovered automatically). The interest in process mining is rising as is reflected by the growing numbers of publications, citations and commercial tools (Disco, ARIS PPM, QPR, Celonis, SNP, minit, myInvenio, Perceptive, etc.). In the academic world, ProM is the de-facto standard ( and research groups all over the world have contributed to the hundreds of ProM plug-ins available.

Understanding the evolution of customer behavior during their journey requires a temporal analysis of the upgrading events. Stream process mining of the generated models guarantees observing the evolution of the interleaving sub-processes and detecting interesting concept drifts. Stream mining is a recent subject in data mining. The task of streaming process discovery includes, among other things, working on potentially unbounded event logs. Few algorithms have been developed, that are able to perform process discovery in a single pass over the data. This master project will explore applying several streaming process mining and stream data mining techniques on the event data in order to answer the following questions:

  1. What happened? Which usage, extension and upgrade patterns do we have? How did they evolve over time? Which concept drifts exist?
  2. Why did happen? Will any, and if so which, customer segmentation help in explaining the answers to (1)? Will splitting interleaving sub-process better relate these answers to e.g. some timing aspects?
  3. What will happen? Can we predict next customer usages, extensions and/or upgrades based on our previous learned models?
  4. What is the best that could happen? How can we use our prediction knowledge in recommending certain controllable actions such that Customer Lifetime Value (CLV) (or other KPIs) can be maximized?


For more information, contact Marwan Hassani.