SenseML : a platform for constructing IOT data pipelines

Abstract

In this thesis, we present SenseML. SenseML is a general-purpose platform that enables users to transform sensor data from the IOT domain into a machine learning-ready format - what we call an attribute time series. It is a cloud-based platform that can process signals using user-specified functions. It offers users immense flexibility in integrating functions for transforming the data, while also providing parallel execution as a service. In addition, we enable users to contribute to the framework by submitting domain-specific signal processing functions. Such contributions are integrated into the platform and are then part of the library, available for others to use. We used the platform to generate 19 attribute time series for 9655 urban sound signals. To generate these time series, the platform did 32 million computations in approximately 140 minutes.