14.1. Introduction

Data science has long been prevalent in all areas of science in this digital era. Data science is an interdisciplinary field that fuses science and technologies by using algorithms, tasks and devices to extract usable data from raw unstructured data. The extracted data is applied to various domains to gain insights from the data acquired to refine the required data through efficient searches. George and Groza (2019) introduced a concept of the extract-transform-load (ETL) that used graduate attributes in the common form of places and details. Separate attention was given to the transformation procedures that helped to get two different reports as final results. The reports were the graduate attribute report per cohort (GAR/C) and course progression report per cohort (CPR/C). The GAR/C accessed each attribute based on the average and the CPR/C showed the tracking information based on the achievement made by the students in their program. The reports were generated synchronously at the same time to enhance the ETL efficiency. The model paved the way for the integrated assessment of the database based on the granularity of the ETL. Sulaiman et al. (2019) came up with incorporating information and communication technologies (ICT) in the power industry by applying modernization concepts to it. That field is the smart power grid, which integrates all the smart meters used in that grid. It collects an enormous amount of data from these smart meters and processes the huge data in the centralized servers to control the entire grid, making it easier to control and observe from a remote location. In this work, the smart meters act as a backbone of this smart grid. The data science is used to process this huge data, thereby benefitting the user and also the energy supplier in that domain of engineering.

Ghosh and Grolinger (2019) investigated the merging of cloud and edge computing for IoT data analytics and came up with a deep learning (DL)-based system to define the data processing along with ML concepts. The encoder is widely used in all the nodes to reduce data congestion while reading data from the sensors. These kinds of reduced data from the devices make the big data feasible to the application. This data has been used directly by the ML algorithm to expand the original features with the help of a decoder present in the auto-encoder module. McHann (2013) developed a strategy of collecting data from all the nodes at present and processing it further at a later stage. But this method involves data to be stored in lot of storage devices. The large volumes of data have to be stored in the cloud to perform ML at later stages. As the data storage capacity increases, the need for technology infrastructure increases, skill sets to work in that infrastructure increase and it is also expensive in terms of time and budget (Bhuiyan et al. 2017). This chapter has been organized as follows. Section 14.2 elaborates on the machine learning models in materials science and its application in Electronic Engineering. It also discusses data acquisition by supervised learning and outlines accessing data repositories and the data storage, respectively. This section describes the comparison of the predicted and the actual values. Section 14.3 illustrates the application of machine learning in electronic engineering. Finally, Section 14.4 concludes the work and recommends the future aspects.

14.2. Machine learning (ML) in electronic engineering

Machine learning (ML) has a wide range of applications in the semiconductor industry. ML in this industry has two different phases of application such as:

• –training phase – algorithm to get results;
• –inference phase – response to stimuli.

The application of ML in semiconductors includes learning from the data. The data has been acquired from the databases or clouds called repositories. Numerous researchers around the globe perform their simulations and experiments and provide their valuable results in the common cloud to prove their integrated research (Karmawijaya et al. 2019; Naveen Balaji et al. 2019; Balaji et al. 2020; Malini et al. 2020). By accessing the data repositories, the ML user can retrieve usable data by using efficient algorithms and refining the model. The model can be trained in this environment by undergoing several searches and improved by the feedback given (Malini et al. 2020).

The various procedural steps in ML are as follows:

• –acquiring data from the repository;
• –handling acquired data by algorithms;
• –supervised learning by predictive models;
• –unsupervised learning through patterns that exist between I/O;
• –experimental procedures design for materials science;
• –cyberinfrastructure – data platforms.

Data acquisition is the initial step in ML where data has been extracted from the data repositories based on the user’s search (Malini et al. 2019). The next step is learning from the data based on mathematical applications such as correlation and regression. Supervised learning (SL) is the method of backtracking the inputs from the outputs, thereby establishing the relationship between the input and output pairs (Casula et al. 2019; Gowthaman and Srivastava 2021a, 2021b; https://www.wolframalpha.com). SL has been the most prevalent research area in the ML platform to refine the model created. Unsupervised learning (UL) is the method of creating the algorithm to study the pattern behaviors that exist between the input and output. The UL is the most time-consuming process that needs a revision of the pattern and creates the model (Kampker et al. 2018). The design of experimental procedures plays a major role in the design of the ML model. This needs a lot of procedures and algorithms to be used in the ML model (Moradi et al. 2020). Hence, cyberinfrastructure comes into existence. The famous cyberinfrastructures are Citrine, The Materials Project, Wolfram Alpha, KIM and Materials Data Facility (MDF) for materials science research (Karpatne et al. 2017; Tanifuji et al. 2019; Liu and Shur 2020; Gowthaman and Srivastava 2021c).

The flow of the ML platform for the materials science project has been illustrated in Figure 14.1. The existing data is present in the data repositories submitted by researchers around the globe. The predictive model suggests the test run of the search of the properties of the particular material (Kampker et al. 2018). The search results have been shared with the information acquisition system for further processing. The information acquisition system collects the search results from above and sends the data packets for verification based on the query made on the web front of the cyberinfrastructure (Liu and Shur 2020). The query has been recorded in the database of the query information source register. The next step of this process is the verification of the goals given to the results obtained (Karpatne et al. 2017).

14.2.1. Data acquisition

Data acquisition is the most important aspect of data analytics in the materials science domain. The individual researcher has to devise a platform to make sure the data acquired is legit and correct as per the requirement (Casula et al. 2019). Every individual researcher has to collaborate with researchers around the world to make an integrated search based on the big data (McHann 2013; Moradi et al. 2020). The next step is the extraction of the result with the publication metric from the open repositories. The cyberinfrastructures are the larger databases of information from many researchers. ML has been used to extract the data required efficiently.

The application program interface (API) is a set of rules or even algorithms to extract the required data from larger datasets (Hack and Papka 2015). The API is widely used by advanced ML users. Beginners always uses a web application function before the API to train the model. The efficient API model is supported by the web function created to enhance the search.

14.3. Electronic engineering applications – data science

The physical vapor deposition method to create the elemental compounds using various elements has been illustrated in Figure 14.3. The marked items are the compounds which have been used in electronic engineering and applied to the DG MOSFETs to get rid of SCEs. The compounds are as follows lanthanum, gallium, aluminum and hafnium (Mohammadi et al. 2018). Their properties have been analyzed for effective usage in electronic applications. The inkjet method of forming the same compounds has been portrayed in Figure 14.4 for further analysis. These elements have been displayed for the number of material level analyses to create/deposit the material (Singh et al. 2017; Singh and Srivastava 2018; Gowthaman and Srivastava 2021e).

These methods highly depend on the ionic packaging value of the particular material. Using ML, the time taken to create the material has been reduced drastically (Chen et al. 2018; Gowthaman and Srivastava 2021e).

This work mainly focuses on the easy storage of the data and its efficient retrieval on demand. This concentrates on the large data volumes of databases called cyber infrastructures (Feng et al. 2019). The data recovery projected in this work had been submissive to the conventional method of data retrieval. The comparison of the previous researches results in good agreement with the novel data retrieval technique. The confidentiality and the security of the user’s data was ensured by normalization of the raw data. The unauthorized user cannot determine the type of data they visualize since it is in a normalized form. Hence, normalizations of the data had given enormous capability of security and confidentiality in the cloud-based data query. The electronic engineering field has been enhanced by the usage of normalization and other statistic modeling of the raw data in order to process it. ML reduces the data processing and data storage compared to non-ML-based statistical models.

14.4. Conclusion and future work

The enhanced database architecture and the data storage facilities facilitate the access of databases through larger query and its assessment. The big data was introduced in this work to reduce human work and data analysis. The normalization of the data helps the user to create a detailed analysis in terms of the processed data. The idea of ML can be further improved in theoretical form to apply statistics models in the raw data to perform quicker processing. Training is not required at the data user end since it uses automated processing of raw data and reporting of the additional queries.