Machine Learning for Sensing Applications: A Tutorial
Vahideh Shirmohammadli, Behraad Bahreyni
Abstract
The developments in microsensor fabrication over the past few decades have contributed to the availability of a wide range of sensors with varying degrees of performance and cost. Many of the recent waves of technological developments such as the Internet-of-Things or wearables rely on such sensors. With the increasing availability of on-board and remote computing power, the trend is to go beyond the simple quantification of events and (re)create context from sensor data using statistical signal processing, or as commonly known, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">machine learning</i> . Within the scope of this tutorial, we highlight the applications of machine learning in sensing and introduce the fundamental stages for creating data-driven models based on simple machine learning algorithms. We focus on algorithms that are simple to implement, provide accurate results, and yet remain <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">understandable</i> to the human developer. The ability to follow how a data-driven model functions is essential in many engineering applications where a trade-off between accuracy and reliability is often acceptable. We provide case studies that utilize the presented material to solve different real-life applications. These examples demonstrate the importance of choosing appropriate features, selecting algorithms, and finally, a study on figuring out the environmental conditions from sensor data.