Internet of Things Enabled Machine Learning-Based Smart Systems: A Bird’s Eye View ^†

Abstract

Machine learning (ML) helps the Internet of Things (IoT) become widely used by automatically identifying data patterns and extracting important insights from the vast pool of observed data. To efficiently serve corporations, governments, and individual consumers, the Internet of Things (IoT) needs machine learning (ML). The IoT gathers environmental data and automates decision-making using sophisticated methods based on human judgement. Data, application, and industry perspectives are used to organise and assess machine learning–IoT literature. We discuss how machine learning and the Internet of Things can make our surroundings smarter by reviewing relevant research. Our analysis includes many cutting-edge methods. We also discuss pandemic control, networked-enabled cars, distributed computing, trivial deep learning, and the Internet of Things. Technological, personal, commercial, and societal concerns face the Internet of Things. Learning how to use the IoT can improve society’s well-being and longevity. We also examine a case study to find comparative results among various machine learning methods integrated with the IoT.

1. Introduction

IoT can change global dynamics like electricity did. Xu and colleagues provide a comprehensive assessment of current IoT application categories, R&D patterns, and industry challenges. Their work improves industrial IoT understanding. Data are like electrons in IoT. This section analyses IoT machine learning breakthroughs and categorises IoT applications by source.

According to the UN, over 50% of the urban population requires better jobs, education, healthcare, and living standards [1]. The National Intelligence Council (NIC) listed the IoT as one of the technologies that could change U.S. power in 2008. The IoT enables Mark Weiser’s ubiquitous computing concept [2]. The IoT is no longer a buzzword; it connects the analogue and digital worlds and alters our perception of the universe. Big tech companies have invested heavily in hardware and software, statistic and transformation insights, and market analytics businesses, and 25–30 billion devices are forecasted to be connected to the IoT by 2030 [3]. Figure 1 shows the Internet of Things today. Europe leads the U.S. in early IoT adoption, while South Korea has the most linked items.

This research has four sections. Section 1 examines IoT data. Section 2 discusses IOT use scenarios. Section 3 discusses IoT–ML integration and future themes where it will be important. Section 4 concludes the paper.

2. IOT Use Cases

2.1. Smart Grids

Smart grids and IoT technology can improve electricity management. Randal Bryant and colleagues found that machine learning will improve smart grid efficacy significantly. Maximising power grid resilience can help customers and power providers construct better power infrastructure to solve future concerns [4]. Figure 2 shows that different data from multiple IoT applications can provide good predictions.

2.2. Intelligent Transport

Urban regions face complex issues like traffic congestion, pollution, and economic losses from traffic disruptions and vehicle collisions. A machine learning-enabled IoT strategy can generate value from networked data, improving services and speeding up innovation [5]. Autonomous vehicles (AVs) could overhaul transportation. Autonomous vehicles use machine learning (ML) and are projected to become AI driven. Machine learning algorithms recognise and classify mobile and immovable items. Alam et al. [6] suggested using deep learning and decision fusion to recognise driving scene elements.

2.3. Smart Homes

Home uses multiagent systems and probabilistic MLA to make the home environmentally friendly [7] to maximise resident comfort and reduce running costs. An automated technique uses user service feedback to start the modelling process. Human activity recognition (HAR) is now considered essential for smart home automation. In a study on human activity recognition (HAR), SVM and evidence-theoretic KNN classifiers outperformed probabilistic artificial neural networks (ANNs), K-nearest neighbours (KNNs), and naive Bayes (NB) classifiers [8]. Taiwo and colleagues [9] introduced a deep learning-based framework for motion recognition by analysing movement rhythms. The model improved domestic power efficiency. However, deep learning is computationally expensive.

2.4. Smart Health

Dawadi and coworkers researched automated cognitive health evaluation using machine learning. Supervised and unsupervised machine learning models are used to classify activity performance and cognitive evaluations [10]. Cognitive systems can perceive, deduce, and learn, making discovery easier and research more efficient.

2.5. Smart Logistics

IoT applications across numerous industries are growing and evolving. Supply-chain management (SCM) and the IoT influence manufacturing. Ellis et al. explained how IoT-enabled analytic tools could revolutionise supply chain management [11]. As shown in reference [12], social media analysis can detect dangers and unwanted situations. Support vector machines (SVMs), naive Bayes (NB), and AdaBoost MLAs are used to pick features and classify the relevant textual content of tweets.

2.5.1. Environmental Control

Shafi et al. [13] used deep neural networks to identify water pollution. Forecasting and prediction use the application of linear regression, random forest, and XGBoost models. Elvitigala et al. [14] calculated pollutant gas concentrations using linear regression. Smart cities can optimise water, land, and air pollution control with IoT and machine intelligence. This can make society safer and healthier, boosting the economy.

2.5.2. Disaster Management

IoT infrastructure could speed up disaster response. Ray et al. [15] examined the IoT paradigm and machine learning-based data analytics in disaster management. IoT infrastructure can help predict forest fires, which is a critical application. Forest fires spread quickly; therefore, the reaction time must be decreased.

2.5.3. Smart Access and Security

The Internet of Things automatically gathers sensitive data and sends them to apps. It is hard to protect the privacy and safety of so many people. Access control is recommended to limit who can see and use data and information [16]. Hussain et al. recently looked at threats, solutions, and problems with IoT security. [17]. The writers suggest using machine learning and deep learning, which are usually used to build smart systems, to protect the Internet of Things.

2.5.4. Pandemic Management

IoT infrastructure and smart systems can help with lockdowns, social distancing, contact tracing, healthcare monitoring, pre-screening, remote meetings, and universal access [18]. The IoT could help reach healthcare goals by giving the public access to virtual healthcare tools and telemedicine [18]. The IoT and AI were used to calm down the staff at the Wuhan Smart Field Hospital. IoT sensors can keep track of infection rates by connecting human nodes [18].

2.5.5. Connected Autonomous Vehicles

IoT will connect autonomous vehicles (AVs) and improve driving cognition. Connected autonomous vehicles are still in development, and their success depends on factors including adoption and price [19]. Pedestrian recognition, intersection navigation, communication, collision prevention, and security are also issues. Alam et al. pioneered this research with TAAWUN [19]. Predictive and connected car data improve driving-scene knowledge. TAAWUN could use IoT infrastructure to discover data for predictive applications in the future. AV trials have resulted in fatalities [20].

2.6. Data-Handling Mechanism in IOT

IoT sensor networks can help make reliable judgements if vast amounts of data are fused and evaluated quickly. The Internet of Things would struggle with widespread settings. To solve this problem, several sensors and smart algorithms are needed.

IoT sensors capture environmental data and send them to a “smart” application, where machine learning algorithms create forecasts and projections. Learning also trains AI to make decisions. Later, actuators implement the choice. Sensor-equipped devices are everywhere. Data from the Internet of Things must be saved, analysed, and archived. IoT infrastructure is still underdeveloped in advanced economies. Moraiset et al. [21] classified IoT data into 19 popular categories. Real-time data floods connected devices. Self-driving car data alone may be massive. For archiving autonomous vehicle data, the cloud may be more adaptive, scalable, and pervasive. Cloud data cannot allow real-time analytics. This emphasises edge and fog IoT data storage [22].

2.7. Outlier Detection

Outliers are abnormal data trends. Low-cost, low-quality sensors; ambient influences; electronic inferences; and data connection problems make outlier observations likely in dense sensor systems like the IoT [23]. MLAs find outliers rather than remove or replace them to gain useful insights from data. MLAs are increasingly used for knowledge acquisition and anomaly detection to improve data quality [23]. WSNs, a subset of the Internet of Things, have studied outlier concerns [23]. Figure 3 illustrates this further.

2.8. Data Processing

The IoT is integrating low-cost sensors with distributed intelligence, changing how we perceive and interact with the world. This combination will generate massive data with valuable insights. Data quality is compromised by imperfect sensing devices. MLAs must pre-process IoT data to extract meaningful information [24]. Figure 4 shows three detected data issues: outliers, missing values, and feature selection.

2.8.1. Handling Missing Data

Data sensing in the IoT ecosystem relies on sensors and RFID devices. Sensors are unreliable measurement tools [24]. IoT apps may have missing values. Synchronisation problems, faulty wireless transmissions, sensor malfunction, power outages, and bad weather cause missing values [24]. Data imputation and deletion of missing data instances are the two main ways to handle missing values in IoT data [24]. Data imputation methods have been developed in natural sciences, census surveys, wireless sensor networks (WSNs), robotics, and scientific applications.

One of the most popular machine learning algorithms (MLAs) is K-nearest neighbour (KNN). KNN-based data imputation is also popular [25]. SVM, a more complex supervised learning algorithm than KNN, is widely used for data imputation. Studies show that they efficiently handle linear and nonlinear data [26]. Multi-class difficulties may reduce SVM accuracy. Breiman and Cutler introduced the ensemble-learning random forest (RF) algorithm, which is newer than the support vector machine (SVM) technique. Reference [26] describes the RF MLA method for imputing missing data values via RF proximity.

2.8.2. Data Feature Selection

Finding the most important data in a potentially overwhelming amount of data is becoming more important. High-dimensional data can reduce accuracy, increase computational complexity, increase memory requirements, and make data visualisation harder for machine learning algorithms (MLAs). Many studies have examined how feature selection (FS) affects machine learning (ML).

In high-density sensor installations, IoT data are extensive and often large. Thus, FS methods are essential for accurate data. Surveys on feature selection and dimensionality reduction are enlisted in reference [27]. KNN, SVM, decision tree, AdaBoost, random forest, naïve Bayes, regularisation, relief-F, and entropy evaluation criteria are used to identify a dataset’s most important variables.

2.9. Machine Learning/IoT

ML could be fundamental to the IoT paradigm. IoT design emphasises accessibility [28]. The product and intelligence promote automation, efficiency, productivity, and connectivity [29]. ML’s pattern-finding ability may underpin human-level intellect. Extrapolating these patterns into broader trends helps us understand the world. Information visualisation’s profound impact on the human visual system makes data and insights easier to grasp in automated systems [30]. Information visualisation helps individuals grasp data quickly without having to analyse them. Guinea and Haiti developed an ML-enabled IoT system with a low-cost sensor network to forecast weather [31].

Technology’s power to improve lives and global prosperity will shift with the Internet of Things [32]. The IoT will generate massive amounts of data, but not all of them will be reliable or helpful. These factors impact the robustness, precision, and efficacy of IoT MLAs.

Many little datapoints are generated for this innovative family of algorithms that must train with limited resources [33]. Sensors are not precise or reliable. Identifying outliers and the imputation of missing values are data preparation tasks that must take place before machine learning.

At Zeitgeist 2011, Google’s head scientist, Peter Norvig, famously said, “We don’t have better algorithms than anyone else; we just have more data” [34]. The IoT offers organisations several opportunities and risks, and the ML-enabled IoT could eliminate millions of jobs [35]. IoT-enabled systems use machine learning (ML) to make sense of large volumes of data from IoT devices. A new class of MLAs is required to handle IoT data. Some MLAs use intuition rather than proofs [36]. Government, commercial, and individual IoT applications use ML. Shanthamallu et al. discussed Internet of Things MLAs [37]. Figure 5 shows that MLAs have several IoT uses.

IoT technology is being used by Cisco, Microsoft, Google, IBM, Oracle, and SAP to create new software systems and hardware that will make our surroundings smarter. As Industry 4.0 is put into place, IoT technology will change the way factories work [38]. Microsoft Azure is a tool for completing work in the cloud [39]. Joseph Sirosh, who is the Corporate Vice President of Machine Learning at Microsoft, says that the IoT creates huge amounts of data every day from many sources. Scott Hanselman, the Senior Program Manager for Microsoft Azure, recently showed how the platform can easily connect many different parts and make machine learning easier for IoT applications [40]. IBM Watson focuses on IoT cognitive computing [41,42]. Google’s open-source machine learning (ML) platform TensorFlow (TF) is notable [43]. Recent advances in Amazon Web Services (AWS) IoT show the importance of machine learning (ML) in the IoT. AWS IoT was linked with AML in early 2016 [44].

2.10. Current Developments in the Field

The Internet of Behaviour (IoB) uses technology to track and study human conduct. It involves gathering data from social media, wearable devices, and other digital platforms to understand people’s behaviours [45]. Sensors, actuators, and connections form the IoT. The IoB combines the IoT, intelligence, and behavioural science. The IoB is a natural extension of the IoT. This study focuses on the dynamic nature of human psychology for better product development and promotion [46]. This could change product and service design, marketing, and customer service. Pinochet et al. [47] examined how IoT product components enhance functional and emotional experiences to increase purchase intent. Stary [48] noted that the Internet of Behaviour (IoB)’s choreographic intelligence might transform business and organisations.

2.10.1. Edge/Fog Computing

Edge computing places computational power near data rather than sending them to a cloud [49]. Edge computing includes fog computing, a new architectural model. Fog computing brings the cloud closer to smart items that generate data and actuators that process them. Edge computing includes data transfer, storage, computation, and networking standards [49]. In cloud, fog, and edge computing [50], deep-learning algorithms involve numerous cloud-processing resources. Edge computing has few processing resources, making it unsuitable for deep learning, while fog computing has even fewer. Deep networks must perceive the environment with fewer data. According to recent deep learning developments, researchers recognise that the Internet of Things needs lightweight deep learning algorithms [51].

2.10.2. Difficulties

Based on several IoT literature reviews, we divided problems into technological, individual, business, and societal categories [52]. The following subsections briefly discuss each. Large-scale IoT adoption is still to come. It is used infrequently worldwide. The IoT combines sensors, networking, actuators, and security. Connectivity issues persist. Internet and cell phone connectivity are difficult in poor countries. The IoT platform’s poor cross-platform capabilities has delayed adoption [52]. Due to latency, energy consumption, single points of failure, and security issues, the client–server architecture that dominates the IoT is unsustainable. Civilisation needs prosperity and sustainability. The IoT aids actionable decision-making [52]. Is society ready for the IoT today? Are we ready to adjust our daily cognitive processes? These critical questions will determine IoT service and application adoption.

3. Case Study: Integration of IOT with Machine Learning

Company: Uber, a Global Ride-Sharing Platform

Implementation: Uber uses IoT devices to track the real-time location of both drivers and riders. During periods of high demand, such as rush hours or special events, Uber employs surge pricing to balance supply and demand.

Machine learning application: Machine learning algorithms process data from various sources, including historical ride patterns, traffic conditions, and events, to predict demand surges and determine appropriate surge-pricing multipliers.

Impact: By leveraging machine learning and IoT data, Uber can optimise pricing in real time, ensuring better service availability during peak hours and generating more income for drivers. This approach also encourages drivers to work during busy times, leading to an overall improvement in service efficiency.

Predicting surge pricing in an intelligent transportation system involves utilising historical data, real-time information, and machine learning algorithms to forecast the likelihood of surge events. In summary, the surge pricing prediction process involves collecting diverse data, pre-processing them to ensure accuracy, engineering features like time and supply–demand ratios, selecting suitable machine learning models such as random forest, training the model with cross-validation techniques, integrating it into the transportation system for real-time predictions, continuously updating and retraining the model, deploying it with rigorous monitoring, and establishing a feedback loop for user input. This approach optimises pricing, improves supply–demand balance, and enhances user satisfaction in intelligent transportation systems.

Table 1. Comparing the accuracy of three different models.

Model	Accuracy
Decision tree regressor	0.9505
Random forest regressor	0.9509
Gradient boosting regressor	0.9377

and Figure 6 show the statistics of the output we achieved through implementation of the models.

4. Conclusions

The IoT has moved forward. More places are using IoT apps. People, businesses, and states all want to take advantage of what the IoT can do. How will the IoT learn and think to automate? This is the most important question. Other areas of computer science use ML to understand and act like people already know the answer. Instead of conducting a literature review, this study looked at how important machine learning (ML) is for creating IoT apps. Breakthroughs in IoT ML are based on data, applications, and industries. The ecosystem of the IoT will benefit from our study. The Internet of Behaviour (IoB), pandemic management, edge and fog computing, connected driverless cars, and lightweight deep learning were all looked into, with a focus on machine learning to find cutting-edge and long-lasting answers. The IoT can make our society smarter and more environmentally friendly, but it needs to overcome technical, personal, business, and social hurdles first. We have come to the conclusion that IoT machine learning will continue to focus on current ML methods for the time being. But it may be possible to make an IoT environment that is fully self-sufficient and has intelligence built in. Due to gadget data and processing power, machine learning (ML) will be hard. This book helps people understand what machine learning (ML) means for the Internet of Things (IoT), how it is used, and what its promise is. To collaborate with various machine learning methods integrated with the IoT by using an intelligent transport system dataset, we found that random forest regression yielded the best accuracy compared with decision tree regression and gradient boosting regression.