Exploiting machine learning for intelligent room lighting applications

Research has shown that environment lighting influences the behavior of the employees in an office setting highly, making lighting configuration in an office space crucial. A breakout area may be used by the employees for various activities that need to be supported by different lighting conditions, e.g. informal meetings or personal retreat. The desired lighting conditions depend on user preferences and other contextual data observable in the environment. In this paper, we introduce a new method for building prediction models to provide intelligent lighting in our pilot breakout area. Based on a set of pre-defined features that are expected to have influence on the users´ choice in selecting a desired lighting environment, we introduce a probabilistic model for generating synthetic data. We also discuss and compare the performances of various rule-based classification models on the synthetic data and find `DecisionTable´ to be the most suitable model for our pilot implementation. We study the influence of the training set size (number of samples) on various classification models and the influences of individual features through simulations. We present empirical results based on the synthetic dataset and a roadmap for future research.