Data Mining for Sensor Intelligence: Change Point Detection and Clustering
Fábio Henrique Pereira · Open Repository of the University of Porto (University of Porto) · 2020
In a growing world of sensors, the importance of reducing their failures keeps increasing, particularly in sensors such as fire detection ones, in which a failure may provoke a useless mobilization of firefighting teams or, in a worst case scenario, not trigger an alarm when there is a fire, possibly resulting in many deaths.To better understand the nature of fire sensor data and to further the cause of reducing their failures, we experiment with a new approach: divide the time series captured by the sensors in segments with different distributions and then proceed to cluster them through a shape similarity distance function.This approach gave us a new way of thinking about the nature of the time series and to better understand what may cause the different shapes.Through that approach, we were able to identify multiple predominant shapes present in the data, to describe them and understand their nature, and to look at the characteristics of each cluster, in order to understand why the clusters exist in the first place.