Statistical inference for intelligent lighting : a pilot study

The decision process in the design and implementation of intelligent lighting applications benefits from insights about the data collected and a deep understanding of the relations among its variables. Data analysis using machine learning allows discovery of knowledge for predictive purposes. In this paper, we analyze a dataset collected on a pilot intelligent lighting application (the breakout dataset) using a supervised machine learning based approach. The performance of the learning algorithms is evaluated using two metrics: Classification Accuracy (CA) and Relevance Score (RS). We find that the breakout dataset has a predominant one-tomany relationship, i.e. a given input may have more than one possible output and that RS is an appropriate metric as opposed to the commonly used CA.