Learning analytics for IoE based educational model using deep learning techniques: architecture, challenges and applications

Since the inception of ‘Internet of Everything (IoE)’ technology, the computing paradigm has totally transformed. New tools and techniques are being devised to handle the varied nature of the voluminous big data generated from the ubiquitous IoE devices. IoE is already exhibiting signs of completely transforming the lives of human beings (Jara et al., 2013; Ashton, 2009; Miraz et al., 2015). A typical IoE system is a collection of large number of devices which are wirelessly connected with each other and the base station. These devices are embedded with micro-chip based sensors, actuators and transponders for performing the task of sensing, storing and forwarding the information about themselves and their surroundings (Jara et al., 2013; Ashton, 2009; Miraz et al., 2015). The different devices in the IoE ecosystem are generally of different make and model which usually stores unstructured data. Therefore there is a vital need of standardizing the data captured by these devices so as to convert it into a common standard format. Furthermore, since the data captured by these devices mainly travel through the wireless medium, it becomes necessary to protect the data in order to maintain its confidentiality. The sensor carries large amount of data which when analyzed in a correct way can give better insights that can be useful in taking well informed and precise decisions. Today, deep learning is considered one of the most prominent technique for performing state-of-the-art classification, analysis and predictions (Du & Swamy, 2013; Chapelle et al., 2009; Zhu & Goldberg, 2009; LeCun et al., 2015). This paper presents an IoE based Educational model and discusses the applications, advantages and challenges of using deep learning techniques to develop a learning analytics system by effectively using the IoE big data for taking better and efficient decisions within the educational domain.

Deep learning is a part of machine learning technique based on learning data representations. This learning can be unsupervised, supervised or semi-supervised (Du & Swamy, 2013; Chapelle et al., 2009; Zhu & Goldberg, 2009; LeCun et al., 2015). In other words we can say that, any deep learning model learns from the experience with minimal external interference. In a typical deep learning system, a computer model is created which is capable of performing the task of classification on the basis of input images, audios and videos. The training of these models is conducted using deep neural network architecture and huge datasets which are generally labelled (Du & Swamy, 2013; Chapelle et al., 2009). Figures 1 and 2 presents the traditional machine learning approach and deep learning approach respectively for performing the task of classification.

This section presents some of the related work in brief. The authors in (Warburton, 2003) reviewed the factors affecting the process of deep learning for performing learning analytics and suggested ways to adopt deep learning strategies among students. The authors of (Luna Scott, 2015) talked about the new pedagogies for the twenty-first century wherein they showed the importance of project based and problem based learning techniques. The authors of (Gul et al., 2017) presented a review work wherein they discussed about the challenges and future directions of IoT in the field of education. The researchers in (Chotitham et al., 2014) discussed the effect of deep learning on the achievements of the students of “Chulalongkorn University”. Golino et al. (Golino et al., 2014) discussed about combining the psychometric and machine learning techniques and its implications to predict the academic achievements of the students. The authors of (Zhang et al., 2016) talked about incorporating “Deep belief Networks (DBN)” and feature engineering to solve the task of “automatic short answer grading (ASAG)” system. Smith et al. (Smith et al., 2015) introduced diagrammatic student model to predict the attainment of concept on the basis of the sketches using deep learning concepts. The researcher of (Nouby & Alkhazali, 2017) investigated the effect of blended learning environment on the attainment of the students of “Arabian Gulf University”. The author in (Rodriguez, 2009) studied the impact of academic “self-concept” and “outcome-expectations” for choosing the learning strategy considering the academic achievement as a primary goal. The Hewlett Foundation (VanderArk & Schneider, 2012) suggested that deeper learning helps in enabling the students to be proficient in the core academics, think critically, work in groups and promotes self directed learning. Angel Fidalgo et al. (Fidalgo et al., 2014) developed a learning Analytics system to analyze the intra group interactions in order to evaluate the work competence of the groups in a university education system. The authors in (Greller & Drachsler, 2012) proposed the visualization of twitter usage by the student for performing the learning analysis using R programming. The researchers in (Dyckhoff et al., 2012) discussed about a generic design framework for establishing a learning analytics system in order to promote quality teaching and learning. They also discussed the various limitations of the learning analytics system. The researchers in (Poon et al., 2017) talked about a toolkit name “eLAT” for performing the task of learning analytics. The “eLAT” is capable of graphically indicating the learner behaviour, assessment and feedback in order to facilitate inclusive teaching and suggests improvements. Poon L K M et al. in (Uskov et al., 2017) discussed about using the logs of LMS to analyse learning capabilities and achievements of online learners using data mining and visualization techniques. Uskov V.L et al. in (Kovanović et al., 2017) provided a detailed analysis and conceptual design of smart learning analytics for smart university. The author in (Linda, 2014) presented a new model for performance analysis of the students using “automated learning analytics techniques”. Their model claimed to provide real time cognitive development monitoring and feedback mechanism for the students. Linda A. (Papamitsiou & Economides, 2016) talked about the application of predictive healthcare analytics. The paper provided seven ways in which predictive analytics can improve healthcare. The researchers in (Rienties et al., 2017) suggested a theoretical framework for the assessment mechanism of the learners by the instructors. The authors in (Xhakaj et al., 2017) discussed about the effects of exploiting learning analytics model on the “attitude, behaviour and cognition” of the learners. They further discussed the usage of (LA-IEF) at the “open university UK”. Xhakaj F. Et al. (Akhtar et al., 2017) analyzed the usage of LUNA (“a dashboard prototype”) and its effects on students and teachers. Akhtar S et al. (Healion et al., 2017) reported the usage of “Computer Support Collaborative Learning environment” in teaching CAD in laboratories. The researchers in (Atherton et al., 2017) discussed the importance of tracing the physical movements to be incorporated in the learning analytics in order to achieve better understanding of in-class activities. Mirella Atherton et al. (Tempelaar et al., 2018) conducted the study aimed at finding the correlation between the online study approach and its positive effects on attainments of the learners. Dirk Tempelar et al. (Ravi et al., 2017) provided an empirical study aimed at showcasing how “learning decomposition data” can be used to deliver better “learning interventions”. D Ravi et al. (Walker, 2012) proposed a methodology using the deep learning techniques to capture sensor data and perform analysis to predict real time activity classification. Furthermore in order to refine the approach, spectral domain pre-processing was incorporated in their approach.

In the proposed architecture, the stakeholders (students, teachers, staff members, management members, parents etc) and other important entities like Library, Entry-Exit Gates, Canteens, Auditoriums, Laboratories, Washrooms, Classrooms, Gymnasiums etc. within the educational system premises are embedded with several microchip based wireless sensors, actuators and transponders using wearable and or fixed devices. The sensors in these devices sense and capture the information about themselves and their surrounding environment and send it to the base station (or sink) for further processing. These sensors produce a huge amount of data which travels from source to destination and vice versa using a wireless medium, therefore in order to secure this data, we propose to encrypt the data using Twofish Cryptographic technique (Whiting & Schneier, 1998; Schneier et al., 1998; Schneier et al., 1999). Furthermore to ensure optimal use of bandwidth and faster data transfer, the concept of Software Defined Networking (SDN) (Xia et al., 2015; Braun & Menth, 2014; Kreutz et al., 2015) has been incorporated in our approach. Finally, in order to perform better analytics and predictions using the data received from these sensors, deep learning techniques have been used. Figure 3 presents the overview of the architecture of the proposed approach.

The proposed system can be adopted to exploit the advantages of the latest innovations and technologies in an IoE ecosystem tailored for the educational domain. Some of these are presented below (Ark, 2015; Brownlee, 2016; Kolb & Kolb, 2005; Baeten et al., 2010).

The challenges of the proposed architecture can be among the following (Jara et al., 2013; Ashton, 2009; Dyckhoff et al., 2012; Madaan et al., 2017):

With its ubiquitous nature and scalable architecture, IoE is finding its applications in almost every domain of human life. Deep learning techniques can be helpful in attaining deeper insights about the data and applications which in turns can be used for creating better software, tools and systems for the wellbeing of the humans. Apart from learning analytics, the general applications of proposed deep learning based IoE Architecture can be among the following (Yaron, 2017; MathWorks, 2017; Schmidhuber, 2015; Najafabadi et al., 2015):

Deep learning techniques look promising because of their baffling ability to reveal intricate formats among the data and provide novel insights to suggest improvements. This is the reason why they are finding their applications in varied computing domains. From the proposed architecture, it can be concluded that deep learning techniques are well suited for getting a student centric learning ecosystem wherein a student gets the customized learning strategy or approach as per their need and desire for better understanding and longer retention. Similarly the lecture plans and lecture delivery modes can be customized as per the understanding and interests of the students to make the task of teaching and learning more fruitful. Table 1 presented here shows a feature wise comparison of the proposed Learning Analytics based Teaching-Learning approach and conventional Teaching-Learning approach.

It is evident from Table 1 that the learning analytics based approach fares much better as compared to conventional learning approach in terms of holistic progress, cognition, attentiveness, retention and attainment of the learners.

With deep learning techniques as a tool, we can train the system in a much more effective way and have better results as compared to traditional machine learning techniques. For example, we can have a deep learning based voice recognition system wherein the system is trained using deep learning algorithms to have an exclusive voice command control over the various components of the system. This can be used by differently-abled persons to control the equipments, appliances and systems. Such systems can also be used as a password for mobiles, bank lockers, home automation system etc. We can also have domain specific feature enrichment techniques which can be used in stock markets, predictive healthcare systems, state-of-the-art security and authentication mechanisms, enhanced seismic data analytics etc. Furthermore we can create deep learning based healthcare systems for better diagnosis and treatment of the patients. If they are effectively merged with IoE technology, the result will surely be beneficial for the humans at large. Although there are certain limitations and challenges associated in implementing deep learning techniques with IoE big data as discussed in the previous section “Advantages of the Proposed Approach”, however these challenges can be converted into opportunities with the development of new techniques and architectures. In the end we can say that deep learning is still evolving but it surely has great potential to further streamline the process of data classification, analysis and predictions.