Enterprise IoT Modeling: Supervised, Unsupervised, and Reinforcement Learning

Abstract

The Internet of Things (IoT)—the internetworking of physical devices—has been a significant advancement in recent decades and has been the catalyst for several other innovations. New Industrial Internet of Things (IIoT) platforms aim to solve the most complex challenge of manufacturers: consolidating all production systems into a single data model. They are used in smart cities, security and emergencies, environmental applications, energy, healthcare, logistics, industrial control, home automation, agriculture, and animal farming. These objects/devices/appliances can generate, collect, and exchange data without human-to-human or human-to-computer interactions. The IIoT is creating an explosion in structured and unstructured data from a growing army of sensors capable of registering locations, voices, faces, audio, temperature, sentiment, health, and others. Billions of IoT devices are interconnected and a huge volume of data is generated. Every device features automation to assist people in the planning, management, and decision-making of their day-to-day activities. Machine learning (ML) techniques are applied to further enhance the intelligence and capabilities of an application. Many researchers are interested in producing advanced IoT technology, combining ML and IoT Techniques. Through ML, IIoT devices learn to perform tasks such as predication, pattern recognition, classification, and clustering. To provide for a learning process, IoT devices are trained using various algorithms in ML and statistical models to analyze sample data. The various fields of data sets (structured and unstructured data) are characterized by measuring functional parameters. Later, ML algorithms are applied to the data set to find features, provide useful output, identify patterns or make decisions based on the data set, draw inferences from real-time data streams, make their results available to analysts, and embed their results directly in business processes. In ML, the real-time problem is classified by classification, clustering, regression models, and association rules. Based on the learning style, ML algorithms can be categorized as supervised, unsupervised, semi-supervised, and reinforcement learning.