Machine Learning Algorithms for Smart Gas Sensor Arrays

Recently, the growth of technologies has resulted in many advancements in the field of gas sensing, one of which is the concept of an electronic nose (known as e-nose) that mimics the olfactory system in the mammalian nose. The e-nose system is developed from a set of gas sensors; however, the notion of the e-nose becomes complete only when the idea of a machine learning (ML) algorithm is implemented. This is because ML algorithms precisely control the e-nose by analyzing the sensor array output data. The development of ML techniques facilitates the analysis of massive volumes of data generated from sensor arrays in the presence of different analyte gases and environmental factors (temperature, humidity, etc.) and then helps to introduce a smart sensor system for various applications. The recent progress in ML techniques has not only simplified the complexity of data from sensor arrays but also improved the potential of e-nose systems by enabling them to accurately classify and predict the type of analyte gas molecules and their concentration. Modern e-nose systems are substantially superior to animal noses since they can predict gas molecule concentrations and detect odorless gases. Therefore, in addition to focusing on material selection and sensor fabrication, it is critical to understand the progress in ML techniques and their impact on the field of gas sensing. Unfortunately, there are very few articles to explain the studies based on ML algorithms and their potential for developing an e-nose system. Herein, a comprehensive review of the ML algorithms and their role in developing an e-nose system is presented. This chapter begins with a journey of ML algorithms such as supervised, unsupervised, and neural network algorithms that are relevant to developing e-nose and discusses the basic idea of each algorithm. Then subsequent sections provide an overview of the role of different ML algorithms in the e-nose system used for various practical applications, including environmental monitoring, food processing, and disease diagnosis. Finally, an outlook on the challenges in employing ML algorithms in e-nose systems and their current progress is discussed.