target="_blank" rel="nofollow" href="#ulink_bc2d5ad5-95ad-5c8f-9210-750048228851">Figure 2.1 Smart cities main features.
2.2 Role of Machine Learning in Smart Cities
Current advanced technologies in sensors and IoT devices make it essential in leveraging AI, particularly machine learning, to model the data for further application [16, 17]. The IoT devices are considered as the most important and unavoidable parts of smart cities. These devices provide a huge amount of data depending on which applications are going to be used, such as healthcare and transportation applications [18].
IoT technologies have proliferated in many fields, such as smart healthcare [13, 19] and smart home systems like Alexa [20, 21], specifically in urban cities, turning them into smart cities. Thus, the huge usage of IoT technologies plays a pivot role in generating big data that requires solutions to analyze and keep track of smart cities' activities. This big data analysis [22] provides invaluable knowledge to integrate all smart cities' sources like IoT and networks. As smart cities and their data grow, this analysis process may become challenging for future decision making. In Section 2.3, we address this problems and discuss possible solutions that have been proposed thus far. Researchers in [23] proposed a new solution to leverage citizen‐centered big data analysis to apply to smart cities, which is to determine a path for implementation of citizen‐centered big data analysis for the sake of decision making. This solution's main goal is to provide imperative perspectives: data analysis algorithms and urban governance issues [23].
Furthermore, researchers in [24] provided several DL applications in smart cities, such as smart governance, smart urban modeling, smart education, smart transportation, intelligent infrastructures, and smart health solutions. Additionally, the challenges of using DL toward smart cities are also addressed. However, still the problem of decision making in smart cities remains challenging. In Section 2.3, we address the problems and highlight the possible solutions.
2.3 Smart Cities Data Analytics Framework
There are plenty of research studies accomplished in smart cities like [25], which is an intelligent decision computing solution for crowd monitoring. In this section, we dig into smart cities applications, and the foundation of techniques is developed upon by establishing a universal smart cities data analytics framework, which is elaborately depicted in Figure 2.2. This framework has three main sections: data capturing, data analysis, and decision making. We elaborate on each in separate sections.
2.3.1 Data Capturing
Smart cities have been engulfed with too much data that requires the management department to control and monitor the cities, but this department is cost and time inefficient. Hopefully, due to domains (features of smart cities), these data, which are grouped automatically into domains, create particular big data for each domain separately. IoT devices used to gather data for healthcare are completely different from the ones that are developed specially for traffic control (i.e. why we need to categorize the data into groups of domains to create certain technical big data for each domain). According to Figure 2.2, we provide six different samples of sensors and IoT devices, such as smart phones, smart cameras, smart thermometers, smart users with sensors, smart cars, and smart houses. The variations of sensors enable a data center to receive different types, ranges, and values of data from objects. Here, we highlight the current upcoming challenges with associate solutions.
Figure 2.2 Smart cities data analytics framework.
In addition to that, data have proliferated significantly and are produced from heterogeneous sources. Therefore, the types of data vary from video and images to digits or strings and need particular procedures to convert all of the data into single unit measurements. These measurements enable us to run machine learning algorithms and other DL algorithms on the data readily to make optimum decisions.
To handle the heterogeneous problem, data engineering [22] is responsible for managing and analyzing the input data and adding labels to data left unlabeled. This requires experts and time; thus it is not cost and time efficient. Therefore, leveraging big data algorithms helps to tune and analyze the data properly.
2.3.2 Data Analysis
The smart cities promises lead us to an ample proliferation and generation in data from all aspects of the domains and branches. Therefore, such huge amounts of data are at the core of the services generated by the IoT technologies [29]. This section of the framework, data analysis, is imperative because its results lead us to make proper decisions. If this process is not accomplished, the decision made will not be efficient. Thus, a large number of research studies have enhanced the process and yielded better results. In the early era of smart cities, there were only limited data generated every day due to the lack of sensors. Therefore, typical machine learning algorithms were sufficient for data analysis to make a model that can handle the situation and provide enough information to make a decision. However, thanks to technologies, the number of sensors and IoT objects have proliferated, and thus we have huge amounts of data that require big data algorithms like and Hadoop to handle the data [30].
Additionally, due to the huge quantity of data, researchers used DL algorithms especially transfer learning and meta‐learning [7] and some other famous machine learning techniques to learn within reinforcement learning like Q‐learning 31–33 for generating smart systems [34, 35].
2.3.2.1 Big Data Algorithms and Challenges
Due to the big data revolution, the enormous volume of high‐performance computations are unavoidable in such smart cities. Thus, big data algorithms are getting one of the important functioning pieces of smart cities.