10.1.1 Internet of Things

Kevin Ashton (in 1998) firstly introduced IoT, which reflects the impending of Internet and omnipresent computing [5]. Such technical transformation reflects the vision of communication and accessibility. In IoT, a thing is an individual or physical entity also on planet earth which has a unique identifier (UID), expert devices, and the ability to transmit data across a network through a non-communicating useless item and communicating system. IoT includes small items or devices which are also connected to tiny things, anyone can access, and whatever may be Internet part. The objects become communication nodes on the Internet through data transmission, primarily through radio frequency identification tags. Things are not only tangible entities but also electronic entities which conduct certain human as well as environmental activities. Moreover, IoT is a model for hardware and software and also covers connectivity and social dimensions [6]. Specially, a three-layer IoT architecture, with network perception layer, application layer, and perception layer; however, several [7, 8] also added layers like business layer and middleware layer. The fundamental of IoT is the ubiquitous computing foundation, comprised of three components [14]: (a) middleware, (b) hardware, and (c) presentation. As per the authors in [15] and [14], three factors are involved in the IoT environment. IoT architecture, according to the authors in [19], is consists of three layers: network, perception, and application. Figure 10.1 shows the architecture of IoT [19].

The lowermost layer is perception layer in the IoT architecture which gathers environmental data whose main goal is to identify the data after the atmosphere. The data sensing quantity and data collection is completed on perception layer [9]. Detectors, RFID tags, labels with bar code, GPS, and sensor are available on just this layer. Its main objective is to define object/item as well as data collection.

The layer (network) consisting of wireless as well as wired systems gathers data provided by engineering technologies from the perception layer. Network collects the input from the lowermost layer and then transmits it to Internet. This works like network management or center for processing information. It may contain a gateway, having two interfaces wherein one is associated to the sensor network and secondly to Internet.

The network layer then gives data to the middleware layer, the second lowermost layer. Service management and data storage is the main goal of this layer. Its another task includes performing tasks and automatically taking decisions based on the outcome. Next, the result of middleware layer is provided to the application layer [8].

This layer is composed of abstract systems that communicate with the end customer to satisfy their needs. This layer gathers information again from middleware layer and provides worldwide technology solutions. Moreover, the application layer introduces the facts in terms of smart city, health, farming, transport, vehicle tracking, home, and many applications [8].

The business layer is the uppermost layer includes make money from service delivered from other layers. The application layer provides data which is then processed to make it knowledge will be transformed into a meaningful service and new services will be generated from existing services.

In fact, the IoT is heterogeneous. IoT’s dynamics, intelligence, and versatility make it is a big demand innovation but also renders IoT unstable and insecure. There are currently only early applications in the environmental sector, such as online particulate source monitoring and indoor environmental control systems. The EIoT covers the environmental sciences; however, it has many possible applications in environmental analysis, modeling, and management.

10.1.2 Cloud Computing

The latest development in information technology which is cloud computing brings desktop computing to the entire web, and still users should not worry about managing as well as regulating any of devices. In cloud computing terminology, the customer has to tolerate just an expense of using the service, called as, pay-as-you-go. Cloud computing will turn a smartphone into a major center of data. Cloud computing is an advanced type of parallel computing, distributed, and grid [10]. Cloud computing provides four types of services: PaaS (Platform as a Service), SaaS (Software as a Service), IaaS (Infrastructure as a Service), and NaaS (Networks as a Service) [11]. SaaS refers to a service that is available to the user on a cost-as-you-use base operating over the Internet [12]. User need not save, install, and manage the program. Alternatively, only internet connectivity is required for the SaaS service provider to access the service that was reserved in the cloud. PaaS offers a framework for creating software and services, with the all the required tools and facilities in doing so [13]. NaaS offers the virtual network(s) for users.

The Features of Cloud Computing

1. It is really wide-scale. Cloud platform storage now has greater than one billion computers, Microsoft, IBM, Yahoo, Amazon, and IBM and tens of billions of other “cloud” sites.
2. The Virtualization

   The feature allows the user to use a diverse range of terminal procurement applications from every site. Assets demanded from the “internet,” instead of a specific fixed object.

3. Resources Pooling

   This means that now the cloud hosting used a multi-cloud model to drag the computing ability to maintain services to various customers. There really are various assigned and reallocated virtual and physical services which rely on the customer’s demand. In particular, the customer does not have influence or information on the area of the resources provided, but can specify position at a higher level.

4. On-Demand Self-Service

   This refers to one of cloud computing’s effective and vital functionality, because the user can track the server throughput, functionality, and allocated network capacity on a continuous basis. The consumer can control the computational resources with this feature too.

5. Undemanding Maintenance

   The servers which are easily controlled and its latency is precise minor; also, there are no downtimes except in a few cases. Depending upon the requirement cloud computing turns up with advancement by creating it increasingly stronger. The new features are more devices coherent and execute likely than older ones together with the fixed bugs.

6. Large Network Access

   Using mobile devices and an internet access, the consumers can connect the data in cloud or publish it anywhere within the cloud. Such features are offered across the network when retrieved through the internet.

7. High Reliability

   Numerous data in the database are kept in the cloud to prevent data loss and improve the reliability. Cloud computing is realer than using local computers.

   Cloud computing is not about a particular application, this can be built together under guidance with ever-changing applications in the “cloud,” with a “cloud” being able to support various applications going on simultaneously.

8. High Scalability

   The size of “cloud” can be vigorously scalable to achieve software requirements and consumer scale development [18].

10.1.3 Environmental Monitoring

It is the collection of physical world measurements which determine the status and trends of environmental conditions. This is vital to human health security, environmental sustainability, and policy growth. In many fields, data monitoring system was widely used, particularly in the situation of weather station. The authors proposed [3] real-time, local measurements, and automatic weather station. Several environmental factors are measured continuously in their system. The results are shown via Blynk 1 platform in an Android and iOS application. The system is composed of two parts that are located indoor and outdoor. The author showed another weather station at [4]. The hardware device is based Zigbee wireless technology and Arduino board. It monitors the meteorological data including air temperature and barometric pressure.

The integrated use of IoT and cloud will relate to the execution of a high-speed information technology among the wide-range controlling entities along with the sensors that are properly implemented in the region. Several other applications may correspond to consistent and long-lasting analysis of water levels (for ponds, rivers, and wastewater), air vapour pressure, soil temperatures, and other qualities. Figure 10.2 shows the relation between IoT, cloud and environment monitoring.

Some potential examples are detection of intrusion in dark areas, fire-infrared radiation, or animal recognition [16]. Other areas of application of this nature include agricultural knowledge transfer and intelligent monitoring, intelligent crop management, food safety monitoring, effective harvesting, and forest detection and tracking [17].

A cloud-based data service is able to overcome the frequency energy requirements of reduced energy communications sectors and the omnipresent, rapid retrieval to end-user data [16]. In addition, it allows the management and processing of complex events, produced by sensor-streamed real-time data.

Environmental monitoring is essential to protecting the environment and human health. As the population continues to rise, as industrialization and energy utilize continue to grow, and despite significant progress in controlling pollution, prolonged pollution production remains unavoidable. Thus, the need for monitoring of the environment remains as great as always. Continued advancements in monitoring system creation, deployment, and automation are required to improve monitoring software accuracy and cost-effectiveness. Now it is crucial to generate more scientists and researchers with the education and skills needed to develop and operate monitoring devices effectively and to manage monitoring systems.

10.2.1 Challenges and Issues

The main issues for IoT resolutions are considered in smart atmospheres as follows: a. interoperability and standardization; b. adaptation and personalization; and c. entity recognition and virtualization.

1. a. Interoperability and Standardization

   The creation of standards which identify all layers from both the physical to the higher layers defines the definition of IoT strongly. The most of them have practical relevance in smart environments and are continuing to evolve. For example, in Industry 4.0 interconnected production requires interoperability among various machines in the sense of smart: cities, health, farming, and factories. The OPC UA offers stable, open platform, and scalable for effective machine-to-machine (M2M) communication to address this concern. OPC UA typically uses mode of transmission and encoding to confirm connection, like embedded control systems and high-end innovation service surroundings [20]. This compromises structure functionality for event with alarms alerts, but also provides users, clients, and servers with authentication capabilities for maintaining the confidentiality of their contact from a security perspective. Many industrial systems for monitoring and automation as well as production networks are usually time sensitive. For this reason, many networks follow the IEEE-TSN (Time Sensitive Networking) unified standard [21] to guarantee a specific time supply through manufacturing systems. The manufacturing process in other instances, such as smart cities, health, or home, is not always as innovative as with the smart industry, but it is still visible and wealthy. In addition, the presence of specifications validates the self-awareness of the technology, while attesting to the huge set of standards and the reality they often have considerable conflicts.

   From IoT providers’ point of view, lack of standardization and interoperability means that the service providers are tied to and must continue with the IoT platform or software provided by a single vendor, which could later carry the potential risk of higher operating costs, product performance, as well as reliability problems [39]. The inconsistency among various IoT systems is helping to momentarily prevent the ecosystem of the IoT service supplier before the IoT industry grows more mature. In particular, support for heterogeneous interfaces of all multiple devices is very expensive for small businesses.

   Both academia and industry have stressed the significance of the challenge in IoT as interoperability. By standardizing, the factories seek to address the challenges of IoT interoperability as shown in Figure 10.3.

   

2. b. Personalization and Adaptation

   A recent trend toward personalization is triggering increased emphasis for transparency and interoperability, both in luxury and professional smart environments. Sensing and control networks which run well in separation must aspect new challenges in adaptation and networking to open IoT networks to various stakeholders, realms, and technologies that enable customization. A significant problem is in what way to change the approach the algorithms used through the stored data to provide the solution’s user the best possible experience or to someone who takes knowledge from the processing. Two factors are important for long-term adoption and use of emerging innovations: they must provide valuable and validated information which will be fairly easy to set up, recognize, and preserve. For example, smart environments (smart cities, smart homes, and smart manufacturers) or health; all such points are critical in order not to unrestraint the technology. With health or safety, devices are assumed to be abandoned quickly [22] as individuals are motivated by the response styles in at the beginning, but this information is inadequate to motivate long-term usage. For example, if an activity monitoring system is unable to sense the quantity of exercise that one is doing since the threshold that determines it does not ensemble ones use. Hence, during the first few days, such a device will “learn” how the person is behaving. Another example of that is pain. The researcher’s aim is to develop a system which will be capable to observe the daily movement of a senior citizen at house to notice variations in her/his behavior and assume any future health-related issues. It is a key problem on the basis of any of this type of distress or health control, as nobody need not initiate too much warning, but we need not miss a case either. The next important attribute is their subjective ease of use, install and maintenance in a latest technology. Installation and proper maintenance is largely an issue of design and technology. However, for ease of use, the embedded algorithms may play a role. Some research groups (e.g., [23–25]) are heading through this issue, are currently underway, and are in their initial phases. The intention is by using device traces to continuously alter the actions of the system, and what has been called subjective and objective user inputs. Analyzing how society behaves well within smart world in order to recognize these data anomalies is very expensive but important for customer experience and device efficiency. If we want even more use and distribution of smart environments and soundscapes in residential care, these systems have to be easy and person-friendly, not the opposite actually.

3. c. Virtualization and Entity Identification

   IoT systems put all together various aspects and functionalities to create interconnected structures of that complexity made up of several interconnected supports. Entity to be identified is a central aspect for dealing this uncertainty, ensuring successful installation and overseeing of entirely functional organizations, as well as ensuring reliable operation in a flexible and responsive context. This, in addition, indicates the need for universal IoT recognition as well as support services to address and link entity codes to related data about data. IoT objects can integrate broadly heterogeneous types of physical objects, manufactured items and devices, places, individuals and other alive plants and animals, and the physical environment and positions. Identification of entities is a necessary prerequisite for IoT object virtualization, which is called a successful innovation toward the interoperability because of its ability to monitor and configure digital and physical resources across.

4. d. Machine to Machine Communication

   Machine to machine communication can be divided into four major components as shown in Table 10.1: sensor layer collecting data; interaction units/layer relaying the collected information; computing units/layer analyzing the information; and service layers taking action. Due to the enormous future use of sensors, another challenge is sensor technology which needs minimal/even zero efforts to maintain and deploy. As per [26], many internal abundant computing plans have declined due to the implementation of sensors is complex next is battery replacement; another is low power sensor design. Because of the increase in number of internet connected devices, do the privacy and security issues [27]. The connected devices can generate data oceans. As per Cisco, in 2008 or 2009, the amount of artefacts on the internet exceeded the quantity of humans [28], a phenomenon that accelerates annually. Therefore, in the time ahead, the sum of machine made data will be instructions of greater magnitude.

   Service	Computers work for frictionless people and robust people Generic framework for improving ecosystem creativity
   Computation	The answer to the questions is determined before Efficient device distribution and cloud intelligence
   Communication	Zero energy efficiency efforts to link huge, compact populations of stationary and moving devices Complete data privacy and security
   Sensors	Less power, hence no battery shift Zero touch for installing as well as handling systems

5. e. Challenges for Database Management

   The major challenges in this area are the huge infrastructure potential. In particular, the environmental uncertainty makes it difficult to have adequate analytical resources to deal for rising environmental factors. In addition, security problems are also associated, because risks may be originate in data breaches due to potential violations induced by an infected system or flaw in the communication system.

   • Scale, Size, and Indexing

     The magnitude and complexity of the IoT facts would be enormous. Data must be succeeded through liable local rights. Owners of community must determine which resources and data the global network will be made available. Thus, the IoT can perform more on than one level: public and private. Users can access data privately or publically over the public internet. Data quality variations can occur, depending on possession and service level. Progressively trust and credibility mechanisms provide users with information on the quality of information.

   • Query Languages

     Structured Query Language (SQL) which relies on structured data is the most projecting examples. Moreover, there are some suggestions for semi-structured data related to query languages, which is more useful of the data available on Internet [31–37].

   • Heterogeneity and Integration

     In [38], the author regard an absence of scalable ideal for designing as well as deploying muses over a diverse array of omnipresent devices as among the biggest challenges in making the IoT a reality.

6. f. Other Challenges

   IoT has been instrumental in transforming human existence with greater connectivity and functionality through the today’s fast-present Internet networking. Nowadays, IoT becomes more subjective and predictive, combining both the technical and the simulated domain to create a highly personalized, often predictive, culture of today. Besides its promise, IoT is about to address three important issues such as unified interface specifications, privacy, and protection. In addition, the IoT development process would be slow unless those standards are available for smart devices. Failure to address related to data security protection at any IoT joint would not only hamper the growth of IoT but also could also result in lawsuits and social security.

   The role of big data, linked to sustainable development, is greatest evident in the energy sector which is faced with productivity and eco-friendly problems related to carbon emission saving [29]. In addition, the usage of renewable and distributed energy production is a crucial component in decreasing carbon dioxide discharges, because the activity of the energy grid accounts for one quarter of global emissions [30].

   A number of technological obstacles remain, however, which must be resolved previously the complete IoT dream converts realism. High among these issues are paradigms of usage of internet, proof of identity, behavior and performance, operation and heterogeneity, and also protection, trust, and privacy mechanism technologies. Classical handling is a matter of competition, for example, concurrency, transaction processing, inter-process, and state communication.

   The author in [35] classified the IoT paradigm problems into three broader fields which technological challenges, socio-economic challenges, and environmental challenges.

   Researchers have witnessed the introduction of several cloud-based IoT systems to consume and process big data replays generated by sensors. Their primary purpose is usually to provide improved services or information by making use of the insights gathered to end consumers. Designing and implementing such platforms however raises new research challenges. The aggregation, processing, and utilization of complex and varied unbounded sources of observation are challenging in general.

10.2.2 Technologies Used for Designing Cloud-Based Data Analytics

Smart environments require various support and services for managing, storing, and collection of data on processing from various sources. The sources mainly concern the devices used in smart environment. The various techniques used in smart environment are as follows. Table 10.2 depicts technological challenges and architecture and heterogeneity.

10.2.3 Cloud-Based Data Analysis Techniques and Models

This chapter describes the key techniques and model which are being used for designing of applications of cloud-based data analysis. The models existing here are based on NoSQL database management systems, MapReduce, and workflows.

10.2.3.1 MapReduce for Data Analysis

The major challenge in internet-enabled devices is the processing of vast volumes of input data. Google [47] proposed MapReduce program to practice this large quantity of data. Now, the MapReduce has been introduced and has proven as an efficient application of handling large data in various domains such as image processing, blog crawling, financial analysis, data analysis, data mining, social machine learning, banking, bioinformatics, healthcare application, and language modeling. It is proven that MapReduce is recognized as important programming model for environments of cloud computing.

MapReduce focuses on locality of data feature to locate the computation jobs close to the data inputted to optimize performance. MapReduce processes large amounts of data which semi-structured or unstructured using number of machines in distributed or parallel environments in contrast to RDBMS. This feature of MapReduce helps to tolerate machine failures. To analyze large amount of data on multiple machines, MapReduce is commonly used to implement algorithms and frameworks for the scalable data analysis.

10.2.3.1.1 MapReduce Hypothesis

The hypothesis of MapReduce model states that user need to state a Map and a Reduce function and these functions are used in any application that is used to convert input data to output data. A map function forms a pair containing key and value and makes a list of intermediate pairs containing key and value. The intermediate values may have the same intermediate keys which can be then merged by the reduce function.

Steps to be Followed by Any Application in MapReduce as shown in Figure 10.4.

1. A job descriptor specifies which MapReduce job to be executed. The job descriptor contains the information of input data, i.e., its location which can be accessed using a distributed file system.
2. The master initiates a variety of mapper and reducer processes on various machines according to the information obtained from job descriptor. Such as at the same instant, the process which reads input data get initiated and starts to read input data. This input data further get partitions into a set of splits; then, these splits get distributed to different mappers.
3. If the partition data is obtained, each mapper method executes the job descriptor map function to produce a list of intermediate key-value pairs. These pairs get grouped further on the basis of keys.
4. These clustered pairs of keys are evaluated and the same reducer method is applied to the pairs that have the same keys. Therefore, each reducer process performs the reduction function which the job descriptor has described. Reducer process then combines the same keys to generate small set of values.
5. Finally, the results obtained from each reducer process are to form the final output data get delivered after collecting as per the location identified by task descriptor.

In large storage systems of data such as data center and cluster of computers, Distributed File Systems (DFSs) are the largely acceptable results for MapReduce systems to access the data.

10.2.3.1.2 Framework of MapReduce

To analyze big amounts of data and to develop parallel applications, Hadoop is commonly used MapReduce implementation, and this framework is adopted for development of parallel and distributed applications using various computer languages.

The various Hadoop frameworks are as follows:

1. a) Distributed File System of Hadoop (HDFS)
   1. 1. This Distributed File System (DFS) provides automatic recovery tolerance for faults.
   2. 2. This allows heterogeneous operating system and hardware compatibilities.
   3. 3. Hadoop also supports high performance and reliability of the results.
2. b) Hadoop YARN

   This framework is especially for management of scheduling and cluster resource.

3. c) Hadoop Common

   This provides the basic utilities that support Hadoop’s other modules. Via the YARN launch in 2013, Hadoop is transitioning from a batch processing program into a batch processing program framework for running a wide range of data applications, including graph analysis, in-memory, and streaming.

10.2.3.1.3 MapReduce Algorithms

All major data mining algorithms such as Support Vector Machines (SVMs) [48], K-means [49], C4.5 [50], and Apriori [51] have been reconfigured in MapReduce over the past few years. Chu et al. (2007) showed that MapReduce shows a linear acceleration with an increasing number of processors in a variety of learning algorithms such as Naive Bayes, newly built networks, and probabilistic clustering expectation-maximization.

10.2.4 Data Mining Techniques

Numerous methods and algorithms in data mining are being used to explore the knowledge after databases.

• Classification

  Classification is perhaps the most frequently used data mining approach which uses a set of preclassified instances to propose a mechanism capable of classifying the record population at large. The process of classifying data includes learning and sorting. In learning, the classification algorithm analyzes the training samples. The data are used in classification to evaluate the correctness of the rules of classification is test data. If the rules are permissible for precision, then it may be applied to a fresh tuples of data [58].

• Clustering

  The unsupervised grouping, called a clustering, is also considered an exploration of data where there is no provision of labeled data. The principal objective of the clustering technique is to distinct the unlabeled data group into some kind of limited and isolated collection of both natural and unknown data structures. No provision is made to provide accurate characterization of non-observed samples generated by a certain distribution of probability. Generally, clustering has two aspects on the basis of which the following can be categorized:

  • Hard clustering: The same entity may refer to a single cluster in hard clustering.
  • Soft clustering: The same entity will refer to various clusters in this clustering [59].
• Regression

  Approach of regression can also be used for forecasting. Analyzing regression will be used to link the relationship among the variables. Parameters are significant variables that are previously specified; the response variables have been trying to anticipate. A lot of real-world issues are not only assumptions, such as this is very hard to forecast if it relies on several dependent variable having complex interactions. Thus, more diverse techniques are also used to predict future values. Neural networks (NNs) may be in use to construct models for classification as well as regression [60].

• Association Rule

  Association as well as correlation are often used to recognize the products which are regularly used from the huge volume of data. Such type of approach supports companies make definite actions, like catalog design, cross-marketing, and study of consumer behavior [61]. Main goal focus on rules that are related to frequently coexisting products, which are used for cross selling, market basket analysis (MBA), and root cause analysis (RCA). The reason for this is to generate the valuable information that defines connections from a huge amount of data between data objects.

• Neural Networks

  NN is a set of interconnected input components or output components for every relationship and therefore has a volume available. Also, at learning process, by changing weights, it can estimate the right class labels of the input item sets. NNs are then used to extract insights through complex data and then patterns can be extracted using them. They are well designed for inputs and outputs evaluated continuously. NN is the popular mining algorithms that are used in huge data set to detect patterns and trends and are very useful for predicting or predicting conditions [62].

• CURE Algorithm

  CURE algorithm is a hierarchical clustering algorithm which contains data set portioning. A combination of clustering and random sorting is employed for managing a large database. Increasing subset is partially clustered for this algorithm, previously broadly divided from the array of drawn datasets. Instead, selective clusters are clustered once again to establish optimal clusters [63].

• BIRCH (Balanced Iterative Reducing and Clustering Using Hierarchies)

  BIRCH method is a series of Hierarchical clustering algorithms. Since it lowers the amount of input/output activities, it is generally used for especially large databases. Clustering is the data mining algorithm that uses group common items in order to easily classify the data. So, a cluster is a collection of entities and is internally consistent but clearly unlike other cluster objects [64].

• K-means Clustering Algorithm

  This is among the easiest algorithm and unsupervised learning algorithms for very effective resolution of challenging cluster-based issues. This approach uses another simple and easy method to classify a given data set into that set of clusters [65]. K-means–based approach, not having little insight into the relationships between them, is used to combine the various observations that are connected together. A few selected features can be used for real-life objects in such an n-dimensional, where n indicates the total range of attributes used to define the clusters [66]. When completed, the algorithm will pick k-points in the space of the vectors.

10.2.5 Machine Learning

The techniques are expressed with the technique of machine learning. Artificial Neural Networks (ANN), Random Forests (RFs), Decision Trees (DTs), and SVMs are popular ML techniques that are commonly used for agricultural management.

• Artificial Neural Network (ANN): Since the mid-1990s, ANNs are used primarily for Pattern Recognition (PR) and, in agriculture, attributes mapping [67, 68]. ANNs are also a popular method for classification and regression, such as crop characteristics estimates [69], rainfall and temperature values [70], soil properties [71], irrigation support water content [70], fertilization optimization rates [72], and crop [70]. As a reference, a method for determining field properties of soil and soil component variability was presented in [71]. ANNs have proved to be well-accepted and more reliable but the results obtained of forecast cannot really be standardized. Studies refer to ANNs as an effective device for estimating crop harvest, because the relation among variables is complex and unknown [73, 74].
• Advanced Neural Networks: Like the Adaptive Neuro Fuzzy Inference System (ANFIS) and the deep learning (DL) methods (DLM), these have emerged to tackle a range of drawbacks of traditional ANNs [75, 70]. ANFIS eliminates the typical fuzziness of farming activities and is known for its efficient high dimensionality. DLMs have excellent success in generalization and learn much more quickly than traditional ANNs. The impractical method, uncertainty, and computational complexity of the ANNs led to possible ideas that are easier to train, like RFs, DT and supporting vector machines, with huge probable for forecast applications in agriculture [77].
• Support Vector Machines (SVMs): Dissimilar some additional kernel approaches, the training set have positive critical overview capability and are highly adaptable to the noise. The author in [78] suggested a primary prevention method for SVM-based sugar beet diseases, using indices of spectral vegetation. Support vector regression was used to retrieve consistent vegetation features, soil mapping, and estimates, but this requires more computational training [79].
• Decision Trees (DT): These are being used regularly in applications of classification, but specified variables may also be taken from continuous soil variables [76] values. The researcher used a Classification and Regression Tree (CART) method to forecast responses of crop changeability to soil property managements and variations techniques. Gradient boosting is a term commonly employed with DTs, enabling greater flexibility and system is to monitor in data modeling [80].
• A DT is a ML classifier based on the tree’s data structure which can be used for supervised learning with a procedural modeling approach; each internal node is labeled with an input feature, whereas the arcs that connect a node to several others (children) are labeled with an input data condition that defines the downward path from the root node to the root node.to the leaves
• Random Forests (RFs): These are precise to the over fitting habit trees to its testing set and then are widely known in applications with mapping of attributes, ensuring higher efficiency and higher predictive performance [81, 82]. RFs have comparatively little training time which is easy to parameterise.
• Deep Learning (DL): This is being a very efficient method which adds more difficulty (“depth”) to the model by extending classical ANN. The most known use of DL is the classification of images [83]. Such as deep neural networks relate to convolutionary neural networks (CNNs) and are commonly shown in image finding, since it uses a mathematical practice called as convolution to interpret images in non-literal approaches [84]. This allows those networks to recognize partially covered things. Additionally, these complex neural networks are susceptible to over fitting, with a large amount of features and completely connected with dense layers. DL needs large datasets so that the system works well and is acceptable.

10.2.6 Applications

Importantly, with both the aid of researchers, smart environments also received popularity, but these smart environments are among the most important features of smart cities [104]. Water and green spaces, air quality, waste management, pollution control, waste management, energy conservation, and urban trees control are studied in [105–110] to shape a smart environment, collectively. Table 10.6 gives the overview of ML algorithms for smart environment. Figure 10.5 shows the applications of smart environment.

• Smart Farming

  The use of IoT is very crucial in farming. In farming, IoT is often used to monitor soil, to regulate water, to specify fertilizer levels, to track plant production, to identify infection, etc. The detectors, hardware, software, and a few equipment can be monitored. The farmer is able to track the farm and deliver water anywhere across. There are other applications, such as smart agriculture, farm aircraft, livestock tracking, and smart cultivation. It reduces human expense, energy, and effort. It is also known as precision farming.

  There are various factors which influence farming systems, such as climate conditions, soil conditions, crop infection and weed control, and water resources. Data scarcity limits current models’ ability to incorporate important factors and is reliable enough to gain the trust of the users. Machine learning techniques that emphasize on analyzing data often face difficulties even though large amounts of data are available. Table 10.7 describes the strongest as well as limitations of the ML methods that are used in smart farming.

  Over the past decade, the data that is collected on farming from detectors such as crop sensors, hand-held devices or drones has risen exponentially. Presence of spectral, spatial, and time resolution data with high quality will process to detailed and robust representations. Smart farming aim has become the willingness to gather evidence on topsoil and plant unpredictability and react to variance on a fine-scale. Big data use is aimed at supporting this aim but there are numerous challenges.

  Strengthens	Weaknesses
  Do very better with data sources input, e.g., conventional databases with presumed or controlled data	Need tracking of poor quality and inaccurate data Need techniques for data transformation and clustering
  Do not claim theoretical preconceived associations	Make very good data-assumptions Conclude homogeneity although there is variance in the inter and intra field Cannot promise successful results forever
  Enable rational and tailor-made decision taking using: Estimated yield Form of crop, and identification features Estimation of soil products The detection of diseases and weeds Climate and weather predictions	Requires a significant number of field knowledge-spatial and temporal campaigns, e.g., Cannot apply a guiding principle, difficult to cover all relevant factors Training demand which can be technical and time-consuming Expert expertise on demand, e.g., for shaping

• Smart City

  Smart cities represent best of several wealthiest and more powerful and most complex smart environment scenarios [85, 86]. It includes many domains like the environment, finance, connectivity, resources, development, organization, and some others posing a wide range of relevant challenges and involving different stakeholders, including city officials, managers, providers, and residents, with likely contradictory outlines. It is obvious that smart cities are not only a new challenge in itself, but it is the area where the issues and challenges maybe more complex so heterogeneous, and many expectations and concerns must be addressed by the technological developments. From an ICT viewpoint, the innovations are also transversal to all realms and problems, addressing a range of e-tourism scenarios [87], e-health [92], e-culture [88], smart energy [90], e-government [89], smart mobility [91], to name but a few. Together with political issues [94], the complexity of technological challenges and accessible technologies [93] are obstacles that can hinder the smart city creation. Thus, no wonder which the compatibility of technical standards and solutions is of prime significance, particularly in the area of the IoT, which is universally acknowledged as an important technical enabler for smart cities [95].

• Smart Factory and Industry 4.0

  Industry 4.0 is an evolving industry trend which is getting the advantages of enabling artificial intelligence and environments driving technologies [96]. As it is popular in application areas like smart homes and offices to obtain, process, and act on various types of relevant data sources, smart automated production systems can also benefit from these technologies. The introduction of smart technological solutions in the industrial world has caused a digitalization. This new paradigm is often called the Industrial Revolution of the Fourth Generation-Industry 4.0 [97, 98] and the Future Factory [99]. This envisages smart industries wherein the manufacturing-enabled IoT and Cyber-Physical System [100] set the foundation for producing intelligent systems through smart systems and practices. By harnessing emerging advances in M2M communication, sensor technology, [101], and machine learning [102, 103], smart devices can prepare, monitor, and enhance their individual manufacturing method with limited social interference. Improved retrieval to Industrial IoT (IIoT) data [97] can enable commercial applications from anywhere and on any platform, at any time. The data-intensive nature of smart manufacturing processes, in effect, would allow timely, accurate, and comprehensive record paths resulting in an enhanced view of several processes and activities that was previously impossible. A result is that, the physical and digital worlds become heavily interconnected.

• Smart Home and Smart Metering

  Home-based networks are being described like the area wherever consumers primarily behave: cloud and IoT has a widespread applicability in home environments in which the combined integration of interconnected embedded devices and cloud allows typical in-house operations to be automated. Nonetheless, the integration of software with hardware items allows the transformation of ordinary objects into knowledge machines that can display resources through a web interface integrated across the Web. Most smart-home systems include sensor networks that link intelligent equipment to the Internet for remote monitoring of their activity (e.g., monitoring of the power use of devices to enhance power consumption habits [111]) or remote control (e.g., electricity and air conditioning management [112]). In general, smart lighting has features that lead significant attention from the research community [113, 114] electricity accounts to 19% of universal electrical conservation usage, and interpretations for around 6% of whole greenhouse gas discharges [115]: smart high managing systems have shown to save up to 45% of the energy used for lighting [114]. In this case, the cloud is the strongest choice to create scalable applications with just a few program codes, rendering smart home a simplistic job [116], and providing the tools required for tasks outside the reach of native networks [117]. In this sense, many problems need to be addressed when developing applications, which are primarily linked to the shortage of reliability and consistency. Web-enabled home-based devices and also the communication can be uniform [111]. In addition, device acknowledgment routines are required to allow simple discovery of appliances. There are also questions regarding reliability related to devices not accessible always, device detecting error, and QoS variable [116].

Video Surveillance

Surveillance video is becoming a significant safety and protection feature of the cities today. Smart cameras configured with smart video study can track activities in urban cities for safety and security and offer alert system by acquiring suspicious activities. In [117], author has introduced and concentrated on video monitoring by presenting video content that includes early detection of fire incidents, illegal behavior, and crowd estimate and smart parking system. This research is focused on video-analysis machine learning techniques with better output and incident monitoring with warning generation advantages.

Smart Energy and Smart Grid

Smart electricity and power systems are core constituent of developing smart city structures; these are main part of integrated energy growth plans, as they can not only promote the introduction of renewable energy and transport electrification, but also allow new value-added services related to energy. Electric grid now cover every area of cities, but with the extension of their capacities through competence and data transfers, future urbanization process and facilities not solely connected to their “internal technological activity” would be under gird. The move to clean energy and sustainable cities would require substantial capital investment, expressed in a concise way in phase. IoT with cloud computing can be efficiently combined to even deliver smart energy delivery and utilization management in heterogeneous systems in both the local including wide area.

Usually, IoT nodes involved in these systems have resources for sensing, processing, and communication but minimal resources. Therefore, computational activities can be adequately requested from the cloud, in which more detailed and dynamic decisions are being complete. Cloud acceptance leads to increased robustness through supplying self-caring frameworks and allowing customer involvement and cooperative operation, achieving distributed generation, quality of electric energy, and response to demand [118]. Cloud computing enables massive amounts of information and data from multiple sources to be analyzed and stored through broad networks.

Weather Forecasting

The author [119] recorded and analyzed daily higher and lower temperature, humidity as well as rainfall intensity after a climate station over a 10-year timeslot to support decision-making and predict rainfall by farmers on selecting crop and water resources. In [119], the author used NN model which shows considerable ensemble learning ability but also the authors noted a requirement for fast developments in technologies and software to manage vast volumes of data.

The agriculture relationship with changing weather is bilateral. Although farming is heavily influenced by climate and environment, it really is one of the sectors of the economy that is affecting climate change. Smart farming can minimize emissions by specific tracking of inputs to field temporal and spatial requirements. Proper management of soil, water, fertilizer, and insect will dramatically reduce emissions while preserving crops and reducing costs of production. Innovative machine learning methods have proved useful in mimicking complex, nonlinear problems in the fields of ecology, climate, and the environment [120]. Figure 10.6 shows the various healthcare applications and solutions using IoT.

• Healthcare Applications

  Because of the increase in the population and chronic diseases, the healthcare services are in more demand. The scenario would be more dangerous when the healthcare services would be out of reach; the major part of society would be affected and will be prone to the various diseases.

  Recent technology can help to solve this issue by using the applicability of IoT in healthcare services that make it possible to keep healthcare services in pocket and available at lower cost. With the implication of this technology, the routine checkups are now possible at the doorstep or even at patient’s house. The combination of cloud with IoT in healthcare system supports more flexibility for data storage and monitoring.

  

Some health care applications are discussed as follows:

• Glucose Level Detection

  Glucose level detection is possible with aid of sensors implanted in device. Due to metabolic diseases, the level of sugar in blood increases and remains high for long period. This monitoring of blood sugar helps the patient for getting the changing intensity of glucose which can be used to schedule feeding time [122]. The devices supported with sensors provide proper monitoring and sensing of glucose. Still, some significant innovations are demanded in this field.

• Blood Pressure Monitoring

  For supervision of good health, the combination of BP meter with wireless equipment enabled smart mobile goes into a part of BP detecting system [123]. The gadget designed for BP detection relies on the proper functioning of electronic devices. Mostly, the BP gadget body is made with mechanical assembly body with corresponding unit of electronics digital system device incorporating with programmed computing system [124].

• Electrocardiogram Monitoring

  Electrocardiography documents an electrical movement of heart that incorporates the basic pulse estimation and also assures the critical cadence, such as myocardial ischemia, belated QT intermediate study, and multifaceted arrhythmias [125]. The computing methods and techniques assisted with internet facility to electrocardiogram can be used at highest degree [126]. The unusual information related to the cardiopulmonary capacity is consistently detectable [127, 128]. In a condition of integration, the presence of comprehensive recognition software on the basis language analysis of electrocardigram messages in the relevant layer of the internet system tool arranges for the evaluation of data of electrocardiogram graph [129, 130].

• Body Temperature Monitoring

  Observing of body temperature is one of the medicinal facilities provided by combining IOT with mobile application. Using a body temperature sensor built in bit version [131], the concept of a mobile internet computing system is fluctuated. For checking body temperature, the primary framework uses computer program for recording and transmission. Especially in case of infants, it is very important to monitor body temperature during the illness of babies, so to avoid any uncomfortable situation; some wearable sensors can also be used.

• Oxygen Saturation Monitoring

  The gadget designed for continuously monitoring the oxygen saturation of patient which gives the notifications about the wellbeing of patient using computational parameters. This device uses a medical sensor that produces a photoplethsmogram from that we can measure the amount of oxygen saturation and blood volume differences in the muscles. Using hardware filters, this analog waveform has been further analyzed to get Sp02 values and heart rate. Such values are scanned using A2D converters in the ATmega32-based SoC, but are preserved and uploaded in the system continuously.