Education 4.0 in higher education and computer science: A systematic review

Education 4.0 is a recently introduced concept focused on innovation, novelty, use of technology and connections with employment and industry. In particular, in engineering disciplines like computer science (CS) it is essential that educators keep up to date with industry developments. Indeed, how CS educators effectively design and implement innovative teaching and learning deserves more systematic attention. This study aims to catalogue and synthesise learning design approaches to teaching and learning within CS: (1) Which innovative pedagogic approaches are used in teaching of CS? (2) Which approaches align with Education 4.0? (3) What skills and competences do educators require to align CS teaching with Education 4.0? Our systematic literature review (SLR) included CS papers published between 2016 and 2020. Two hundred and thirty‐one studies were identified of which 66 were included in the final phase, which were coded by a multidisciplinary team. The findings indicated that many CS educators included Education 4.0 learning design elements. We found a clear distinctive three‐cluster solution: (1) EDU4 light, (2) project‐based/hands‐on learning and (3) full EDU4 (refer to Reference [7] conceptualisation, while Education 4.0 refers to our own definition [Reference 71]. These findings suggest three broad flavours when designing innovative CS practices, which might help educators align their practice.

decisions by educators impact learners in the last 10 years. Nonetheless, there remains a strong need to 'explore how educators plan, implement, and evaluate learning designs' [51], and how students react and engage with these learning designs.
In particular, there is a paucity of research within the broad engineering discipline of computer science (CS) on how educators design for innovative practice [35,63,76]. Due to the strategic importance of CS in engineering and supporting the current and next generation of students to develop appropriate computing and data skills [30,35,43,63], we specifically want to explore how CS educators are producing and implementing innovative learning designs.
This study aims to catalogue and synthesise learning design approaches to teaching and learning within CS that are aligned with Education 4.0. There are a wide range of definitions of education 4.0, but common elements include a focus on connections to employment and industry, innovation, novelty, technology use [7,43,46,74,93]. Given the contemporary conceptualisation of Education 4.0 and the rapid pace of development within CS, in this systematic literature review, we will review CS literature published in the period 2016-2020. Furthermore, as there is continuous change in technology and CS, we are specifically interested in whether (or not) innovative CS approaches refer to the skills CS educators need to be able to teach in an Education 4.0 manner. Therefore, in this study, we will aim to address what innovative pedagogical approaches have been used to support CS teaching (RQ1), which pedagogical approaches are consistent with Education 4.0 (RQ2) and finally what skills and competences do CS educators require to align their CS teaching with Education 4.0 (RQ3)?

| LITERATURE REVIEW
CS is a broad subject area that covers many disciplines and overlaps with many others. We use the definition provided by the UK Quality Assurance Agency Subject Benchmark Statement on Computing: 'CS provides the necessary knowledge to understand and build computational systems' [69]. The statement goes on to list the main characteristics of CS and notes that, '[g] enerally, these are expressed in the ability to specify, design and write computer programmes' The breadth of the field means it is able to draw on teaching methods from diverse disciplines and is also used as a way of preparing students for a wide range of professions.

| Previous systematic literature reviews on innovative approaches in CS
Recently several SLRs have been published on how CS educators design and implement innovative learning and teaching approaches, for example, [35,63,76,86]. For example, Aničić, Divjak and Arbanas [63] conducted a meta-analysis of 155 papers from 1980 to 2014 of how CS educators align their curriculum to career development and developing employable graduates. The findings indicated that the curriculum designs and implications need to be aligned to the needs of the industry. As argued by Aničić, Divjak, and Arbanas [63] 'the literature indicates a need for innovative approaches in curriculum design and delivery, such as designing competency-based programmes that are not restrained by the traditional semester seat-time model, providing flexible curriculum and minimising the time spent in the classroom, or offering courses on not only how to manage innovation, but also on how to innovate'.
In a review of 157 learning designs implemented at The Open University, Toetenel and Rienties [86] found that the majority of educators primarily used two types of learning activities, namely, assimilative activities (e.g., reading, watching videos and listening to audio) and assessment activities. Often educators combined assimilative, productive (e.g., coding) and assessment activities or, alternatively, assimilative, finding and handling information and communication tasks (e.g., working together with peers). In a more recent study by Garousi et al. [35] 34 papers were analysed how software engineering education was aligned with industrial needs. Their findings indicated that to encourage the development of soft skills, educators need to use real-life projects, implement industryacademia collaboration in the design of education and anticipate future trends. Furthermore, educators need to prepare students to deal with those trends [35].
In this journal, Kocdar, Bozkurt and Goru Dogan [46] recently reviewed how engineering through distance education could potentially support effective learning and Education 4.0 in particular. Using an innovative approach of text mining and social network analysis, the 120 studies identified indicated that primarily distance education in engineering was provided via technology-enhanced engineering education, e-learning and m-learning, as well as virtual and remote labs [46]. While these studies provide important insights into how engineering and CS educators in particular implement learning and teaching, none of these studies specifically focused on, mentioned or included Education 4.0 concepts. Furthermore, none of these reviews specifically looked at the way the respective learning designs were used.

| Education 4.0
This SLR catalogues approaches to teaching and learning within CS that are aligned with Education 4.0. This is a relatively new term-Harkins originally proposed it in 2008 to describe innovation-producing education [40] as opposed to knowledge-producing education. Education 4.0 is related to the view that the current Industry 4.0 is becoming more and more automated, using modern smart technology and the Internet of Things (IoT). Recently, the World Economic Forum [93] positioned a broad range of eight skills to refer to Education 4.0, including global citizenship skills, innovation and creativity, and life-long learning. Other researchers provide more detailed pedagogical descriptions of what kind of learning activities could be present in Education 4.0, including Fisk [33] and later on Hussin [7], leading to nine characteristics associated with EDU 4.0, as illustrated in Figure 1.
The elements identified are all potentially innovative, and currently the Education 4.0 conceptualisation of Hussin [7] is the most cited Education 4.0 paper in Google Scholar. However, Hussin [7]'s focus on students rather than on the broader picture of how innovations are developed and embedded in terms of learning design is a potential caveat. In this study, we propose an alternative definition of Education 4.0 that draws on ideas and descriptions in a range of literature [11,25,43,68,74,83,89,93]. Building on our initial explorative work [71] how European CS teachers built and designed innovative practice, we defined Education 4.0 as an 'approach to learning and teaching that emphasises the development of skills and competences necessary in a modern workplace using up-todate technology. The skills and competences developed may relate directly to the technology, or they may be the softer skills (such as team-working and creativity) that are needed to work effectively in such an environment. The approach involves the use of technology and/or pedagogy that is innovative in the context, and therefore requires flexible and creative approaches to its implementation'. Note that in the remainder of this study, when we use EDU4 we refer to the Hussin [7] conceptualisation, while Education 4.0 refers to our own definition [71].
As evidenced by a range of studies [35,46], being able to design and implement innovative pedagogical approaches requires substantial new and/or updated skills and competences from educators to make use of Education 4.0 approaches. A recurring theme seems to be a shift from CS educators as being a knowledge transmitter to an educator as a facilitator or moderator or consultant of learning [71]. Educators could achieve that by being flexible (adapt to change) [71], supportive, help students to develop ownership of learning, foster an environment where students take risks and share what they do not know about, and where failure is acceptable. This role was often discussed within a flipped classroom implementation [28] that could give control to students to study the teaching material at their own pace and contact the teacher to solve problems and discuss their learning. In such conditions, the teacher is monitoring a student's progress and facilitates understanding through discussions [54]. An increasing number of CS educators have started to implement project-based learning and hands-on experiences in their classroom [4]. However, the specific skills and competencies needed to design, implement and evaluate effective Education 4.0 CS courses has received limited attention.
In our initial work [71] exploring how 20 European CS studies in nine European countries designed and implemented innovative teaching and learning, we found some preliminary evidence that educators indeed design CS practices by incorporating some of the nine EDU4 elements. On average we found 4.10 (out of nine) EDU4 elements in these 20 studies but with substantial variation (SD = 2.10) and range (2)(3)(4)(5)(6)(7)(8)(9). A preliminary cluster analysis suggested three initial clusters of practices among European CS educators, which we initially labelled as EDU 4.0 light (i.e., incorporating one or two EDU 4.0 characteristics, n = 8), Project-based/ hands-on learning (i.e., using project-based/hands-on approaches, n = 6), and Full EDU 4.0, using a broad range of EDU4 characteristics, n = 6). Given the relatively small size of this European data set it was difficult to generalise the findings across the globe, and furthermore it did not allow us to conduct more robust statistical analysis.
Therefore, in this follow-up study, we extend our search beyond Europe to synthesise how CS educators design and implement innovative CS practice (RQ1) in countries across the globe, and whether (or not) these align with Education 4.0 (RQ2). Furthermore, by enlarging our sample from 20 to 66 papers it will allow us to conduct a more in-depth synthesis to explore whether these patterns are common to CS, or specific to Europe. Finally, a specific new element previously unexplored is that we would be keen to identify what skills and competencies might be needed for educators to align their practices with Education 4.0 (RQ3). This leads us to the following research questions: RQ1: What innovative pedagogical approaches have been used to support CS teaching? RQ2: Which pedagogical approaches are consistent with Education 4.0? RQ3: What skills and competences do CS educators require to align their CS teaching with Education 4.0?

| METHODS
In this SRL, we follow recommendations from [46,53,55,70] and build on our initial approach [71]. In total, four research databases were analysed based upon their ranking and coverage of CS research: Science Direct, Scopus, Web of Science and Wiley InterScience. Papers had to be published in English during the 5-year period 2016-2020, thereby increasing the chance that a particular study used a contemporary and innovative pedagogical approach in CS. We used the following search string: 'computer science' AND education AND teaching AND pedagogy AND ('undergraduate' OR 'postgraduate'). Two hundred and thirty-one unique publications were identified across the four databases.

| Coding process
We followed the same coding procedure as in Rienties et al. [71], whereby an interdisciplinary team of academics from CS and educational technology from six European countries analysed the data. Initially, all abstracts were manually screened, and 75 studies were excluded as the focus was on elementary/secondary education, subjects other than CS and/or learners instead of teaching. In the first phase, after 1 h of online training and discussion on the coding scheme, 18 members of the TEACH4EDU project read 156 studies in detail. Members coded on average eight studies (range: 3-11), of which 68 studies were included in the subsequent analysis following three inclusion criteria: (1) Is it an 'innovative' application in a CS course? (2) Does it use technology or pedagogy in an innovative way? (3) Is the innovation evaluated, if so how? As notions of innovation and technology can differ, in the second phase we specifically assigned selected papers to coders from different disciplines and geographical contexts to check whether the proposed approach was indeed within these inclusion criteria.
In the second phase, 17 members of the TEACH4E-DU project participated in another 1-h online follow-up training and discussion of their 20-variable online coding scheme. A new set of studies was randomly assigned to coders to compare to their initial coding in the first phase, thereby ensuring that at least two coders checked and independently coded each 'innovative' pedagogy in CS. While text mining and bibliometric techniques might be useful to identify common trends in literature, given the complexity of how scholars might write about learning design and Education 4.0, in this study we argue that human verification and validation of innovation are required. Thus for RQ2, we adopted the nine key EDU4 characteristics of Hussin [7] and our Education 4.0 definition for human coding of each included paper. For the analysis, we used both the individual EDU4 scores as well as the aggregate score (e.g., Schäfer [56] was coded as EDU4 1 Learning any time and 9 More independent, leading to an aggregate EDU4 score of 2).
For RQ3, coders indicated whether (or not) any specific skills required by educators to support the teaching of CS to students were mentioned. If yes, coders could use a follow-up open text box to add any description and conceptualisation of educator skills. Following this, we recoded and aggregated the skills.
Based on the coding scheme developed from the RQs, an average of four studies (range: 2-10) were coded per coder. A sample from 15 studies was double-coded to show reliable coding (mean Cohen Kappa EDU4 = 0.84). The coders of Phase 1 then checked the codes of the coders in Phase 2, discussed any differences, and agreed on the final coding (average Cohen Kappa EDU4 = 0.93). If a study did not indicate any EDU4 characteristic, we removed it from further analysis, and therefore we ended up with a total of 66 studies, as illustrated in Figure 2.

| Data analyses
The vast majority of studies included referred to undergraduate CS students (79%), followed by a mix of undergraduate and postgraduate students. Five studies did not explicitly mention the specific student population, and one included teachers only. A total of 36% of studies were from the United States, followed by Spain (9%), Brazil (8%) and Germany (6%). Using the GLOBE geocultural regions classification [42], which has clustered countries across 10 broad geo-cultural regions, 47% of studies were conducted by Anglo-Saxon countries, followed by Latin American countries and Latin European (each 12%), Eastern European (8%), Scandinavian and German countries (each 6%), Confucian Asian (5%) and Middle Eastern countries (3%). No studies were identified from African or Southern Asian countries. Using analysis of variance (ANOVA) analyses, no significant differences were found on our key variables and GLOBE, indicating no substantial differences in practices in CS based upon national/geo-cultural regions.
In terms of reporting the findings of RQ1-2, we first explored the overall data, then carried out an exploratory factor analysis (principal component analysis) with direct oblimin rotation to identify a common structure in the EDU characteristics. Mundfrom et al. [56] suggest that a sample size of respondents 40-60 would be appropriate when there is a good level of agreement among items. Perhaps slightly different from normal surveys where participants independently from each other complete a survey, in our study each of the 66 papers were (double) coded by experienced academics/coders, so therefore we argue that an even smaller sample could lead to a reliable factor structure. Multiple factor structures were explored, but a two-factor structure had the best fit. Finally, a k-means cluster analysis was conducted to explore any common patterns in terms of learning designs employed by CS educators. We explored a range of clusters (from 1 to 5), and subsequently analyzed the explained variance using ANOVAs. When more than three clusters were included some of the clusters became too small to be meaningful, while with less than three the explained variance was smaller than those with three clusters. For RQ3, all articles were screened whether (or not) reference was made towards educators' competences and skills to implement a respective innovation. If a study explicitly mentioned this, it was coded and included in an open text box. These open-text boxes were later analyzed by authors C. H. and J. S. to find common patterns.

| RESULTS
In terms of RQ1, 66 studies included at least one EDU4 characteristic [7]. Furthermore, in total 54 articles (80%) were considered to fit our own Education 4.0 definition. However, perhaps surprisingly none of the articles F I G U R E 2 Visualisation of selection of included/excluded studies. explicitly mentioned the terms Education 4.0 or EDU4. As indicated in Figure 3, on average the 66 studies included 4.41 of the nine EDU4 characteristics of Hussin [7], with a substantial variation (SD = 2.30). There seemed to be two peaks in Figure 3, whereby 35% of studies only had two to three EDU4 characteristics, with another peak at seven EDU4 characteristics.
As indicated in Figure 4, the most common EDU4 characteristic was 5) hands-on learning (73%), followed by 9) more independent (67%), 4) project-based learning (61%). Around half of the studies included the characteristic that 1) learning any time/anywhere, while around a third of studies included 7) assessed differently (35%) and 8) student ownership of curriculum (32%). Furthermore, as illustrated by the error bars, there was substantial variation in the 66 CS practices. These findings confirm initial findings reported in Rienties et al. [71].
In terms of RQ2, with our larger data set we were able to explore how the nine EDU4 elements were related to each other. We found a moderately strong correlation (rho = 0.429, p < .01) between the aggregate Hussin [7] and our Education 4.0 definition, with the strongest correlation on the EDU4 characteristic 5 (rho = 0.417, p < .01). The individual EDU4 characteristics were not all directly and significantly correlated. Therefore, an explorative factor analysis was conducted on the data collected, which indicated the existence of two factors with item loads of 0.45 and more. The first component had an eigenvalue of 2.62 (corresponding to 29% of the explained variance), the second component had an eigenvalue of 1.45 (corresponding to 16% of the explained variance).
As indicated in Table 1, EDU4 characteristic 2, 1, 6, 9 and 3 loaded on the first factor, which we will label as 'individual choice and development'. EDU4 characteristic 4, 5 and 7 loaded on the second factor, which we will label as 'intention project-based/hands-on learning'. EDU4 characteristic 8 did not load on any factor. The respective Cronbach Alphas for these two factors were .68 and .62. In other words, educators often combine EDU4 characteristics together when designing and implementing CS courses based upon these two factors.
Follow-up analysis using the k-means cluster method using the 66 instead of 20 studies confirmed the three cluster model initially found in Rienties et al. [71]. As shown in Figure 5 in line with our previous smaller sample of European CS practices, three clusters of studies were identified, which we (re)label as (1) EDU4 light (n = 18), (2) project-based/hands-on learning (n = 22) and (3) full EDU4 (n = 26). With the notable exception of EDU4 characteristic 8, using ANOVAs all EDU4 characteristics were significantly different between the three clusters with large effect sizes, indicating clear unique clusters (p < 01, η² for EDU4 characteristics F I G U R E 3 Histogram of EDU4 concepts identified in the 66 papers (range: 1-9). ranged from 0.204-0.611). In other words, there appeared to be three distinct innovative pedagogical practices present in published work on CS across the globe in the last 5 years.
As indicated in Figure 6, EDU4 light studies (blue circles) mostly had relatively low total EDU4 scores, and often did not include project-based activities. Therefore, most of these studies in Figure 6 were positioned on the F I G U R E 4 EDU4 characteristics present within the 66 papers (with error bars, range: 0 = not included to 1 = included). A total of 73% of papers included 5) hands-on learning, while 32% of papers included 8) students' opinion in updating curriculum.
T A B L E 1 Pattern structure of Factor analysis EDU4 Characteristics (order based upon factor loadings).

2
2) learning will be personalised to individual students 0.769 1) learning can take place anytime anywhere 0.644 6) students will be exposed to data interpretation in which they are required to apply their theoretical knowledge to numbers and use their reasoning skills to make inferences based on logic and trends from given sets of data 0.618 9) students will become more independent in their own learning 0.617 3) students have a choice in determining how they want to learn 0.589 8) students' opinion will be considered in designing and updating the curriculum 4) students will be exposed to more project-based learning 0.883 5) students will be exposed to more hands-on learning through field experience (e.g., internships, mentoring projects, collaborative projects) 0.836 7) students will be assessed differently and the conventional platforms to assess students may become irrelevant or insufficient bottom left. In contrast, while some project-based/handson learning studies (green circles) also had relatively low EDU4 scores, in particular for personalised learning and choice how to learn, they had a strong focus on projectbased and hands-on learning. Therefore, many of these studies are positioned in the middle to top-left quadrant of Figure 6. Finally, studies that were classified as full EDU4 studies (blue triangles) were mostly positioned in the middle and right of Figure 6 indicating these studies used more and even all EDU4 characteristics in their designs. Note that the numbers in Figure 6 refer to the studies discussed below.

| EDU4 light
As indicated in Table 2, in EDU4 light studies educators mostly focused on more independent (61%), learning any time/anywhere (44%), personalised learning (39%), and choice how to learn (39%), but with limited hands-on learning (17%) and no project-based learning (0%). For example, Schäfer [77] introduced the concept of a modern C++ course for students of electrical engineering and CS based on an inverted classroom with attractive IoT hardware. The main goal of the new course was to reduce lecture time in favour of practical learning of students through programming. Burrows and Borowczak [17] explored one group's use of action research and lesson study in three US university-level CS courses to improve undergraduate Engineering student soft skill understandings. The group collaborated to enhance participant engagement by utilising one lesson focused on soft skills over three research lesson iterations in three distinct university semesters. In another study, Park and Kim [60] identified that while assignments on kernel programming are essential parts of operating system courses taught to CS students to provide them a deep understanding of realworld operating systems, often students struggle. In particular, students were routinely flustered by the daunting task of building a practice environment from scratch, and instructors were pressed for time while validating student work that required several kernel installations and reboots. Therefore, Park and Kim [60] provided a cloud-based system that facilitated a high level of accessibility, enabling students to work on their assignments anywhere they want. While each of these studies listed in Table 2

| Project-based/hands-on learning
The second group, which we called project-based learning/hands-on learning, placed a strong emphasis on project-based learning (86%) and hands-on learning (86%), with relatively limited focus on choice how to learn (5%), personalised learning (5%), and learning any time/anywhere (18%), as illustrated in Table 3. For example, Caceffo, Gama and Azevedo [19] assessed the benefits of the use of technology and active learning practices (i.e., Project-Based Learning and Peer Instruction) in the classroom with 25 students to contribute to a more effective and efficient learning environment. Alomari, Ramasamy, Kiper and Potvin [4] described the use of an innovative platform to improve the knowledge of 51 CS students about software testing by providing a set of learning objects and tutorials categorised by difficulty level. This evolved into a collaborative learning environment that included social networking features such as the ability to award virtual points for student social interaction about testing [4].
Another study [19] assessed the benefits of the use of active learning practices teaching algorithms, data structures and programming logic in a CS introductory course with the feedback from two instructors and 24 undergraduate students via interviews and surveys to contribute to a more effective and efficient learning environment. The study indicated that both students and instructors enjoyed the use of new technologies and active learning in the course, but they would like to prioritise two-way communication between students and instructor, and collaboration among students during class.
In another study, Carrascal, del Barrio and Botella [20] proposed an approach for teaching quantum computing which included using classical object-oriented programming to programme a basic quantum simulator. The approach emphasised practical training and was aimed at students who do not necessarily have a deep theoretical grounding in Mathematics. After a mandatory introduction that covered the basics, the course made use of problem-based learning.
As illustrated in these studies as well as in Table 3, there was an extensive focus on project-based and hands-on learning. However, due to the nature of project-based learning in almost all these studies, flexibility in anytime, anywhere, personalisation and learning choices was relatively low.

| Full EDU4
The full EDU4 cluster was strongly focused on hands-on learning (100%), more independent (96%), personalised learning (85%), learning any time/anywhere (77%) and choice how to learn (77%). The lowest EDU4 characteristic was student ownership of curriculum (38%), as illustrated in Table 4, although this was substantially higher than the other two clusters. For example, Pivkina [67] described the experience of using an undergraduate student as a peer learning assistant (PLA), supporting 80 students in three different undergraduate CS courses with 20 students. PLAs held office hours, helped with labs and tutorials and facilitated student group work in class. They were therefore practicing tutoring their peer students. The impact of the experience was measured by comparing student

Gonçalves, von Wangenheim, Hauck and
Goumopoulos, Nicopolitidis, Gavalas and Paschoal, Oliveira, Nakagawa and Souza interactions between PLAs and regular university teaching assistants, showing a preference of students to interact with their PLAs. While that research offers a case for practical learning, it does not offer any choice of learning anytime or anywhere, limiting the experience to the interaction with the PLAs. A similar case was found in Seyam et al. [79] who evaluated whether pair programming (an agile software development practice, used in both industry and education, which enforces a role-based approach to learning new programming concepts) would help 53 students during five sessions in a mobile development course to better understand mobile programming. For the evaluation of the experience, observations and questionnaires were used to show a rich experience where programming for mobile devices goes beyond merely writing code. This study shows a practical hands-on learning case for the students in a real-life software development environment, but students did not have the option to choose the way to learn or personalise their learning experience.
Active learning strategies based on students' practical work combined with continuous feedback (such as the Inspection-based strategy based on doing and reflection) are preferred by students in their education. According to Silva, Polo and Crosby [81] using these active learning strategies was not appropriate at the beginning of a course, as their use could confuse students. Students need to have prior knowledge about the content to use active learning strategies.

| Skills for educators in CS to deliver EDU4
In terms of RQ3, nearly half of the studies (n = 30) reviewed made an explicit reference to skills and competences CS educators should have or develop to align their CS teaching to innovative practice. As indicated in Figure 7, studies that referred to skills and competences of educators on average had higher scores on nearly all EDU4 characteristics, with the notable exception of 6) data interpretation. A follow-up ANOVA analysis indicated significant differences with a medium effect size between studies that did and did not mention skills of educators on 3) choice how to learn (F = 4.758, p < .05, η² = 0.069); 5) hands-on learning (F = 5.689, p < .05, η² = 0.082); 8) including students' opinion in updating curriculum (F = 5.922, p < .05, η² = 0.085); and 9) more independent learning (F = 7.422, p < .01, η² = 0.104).
Furthermore, the aggregate EDU4 score was substantially higher in studies that mentioned skills and competencies of educators (M = 5.30, SD = 2.20, F I G U R E 7 Average reference to skills and competencies of educators and EDU4 in 66 studies (0 = not mentioned, 1 = mentioned). n not mentioned skills and competences of educators = 36, n mentioned = 30. F = 9.304, p < .01, η² = 0.127) relative to those who did not mention those skills (M = 3.67, SD = 2.14). Although no significant differences across the three clusters were found between studies that did or did not mention skills and competencies of educators, 53% of studies who did were part of the full EDU4 cluster, while only 28% of studies who did not were part of that cluster. In other words, those studies that explicitly referred to skills and competencies for educators seemed to be more explicit and innovative in terms of pedagogies and EDU4 elements. An alternative explanation could be that the authors who employed more innovative pedagogical approaches and full EDU4 modes provided more narratives about how educators could effectively support these innovative approaches.
The studies that made an explicit reference to the skills and competencies of HE educators discussed the issue of educators' skills in relation to the implementation and assessment of an innovative learning intervention, which was the main focus of the article. A reference to or discussion of skills and competencies was often presented as an implication of the proposed study rather than being examined as the starting point of a given article. This could be explained by the fact that innovative teaching approaches or interventions are more likely to require educators to develop new skills and competences, and thus such a discussion was seen as very relevant. This observation could explain some of the insights of the quantitative analysis, in particular the observed higher scores on the EDU4 characteristics in studies where skills and competences of educators are discussed.
In terms of the educators' skills discussed in these studies, some of them could be seen as generic, such as the creation of student-centred environments. Others were more concrete, such as the use of specific mobile games in teaching. A recurring theme we identified was the educator as facilitator, moderator or learning consultant [12,36,44] as opposed to a educator controlling or being the centre of the learning process [47]. Educators could achieve that by being flexible (adapting to change) [36], supportive, helping students to develop ownership of learning [47], fostering an environment where students take risks and share what they do not know about and where failure is acceptable [12].
This facilitative role was often discussed within a flipped classroom implementation [47] that could give control to students to study the teaching material at their own pace and contact the educator to solve problems and discuss their learning. In such conditions, the educator was monitoring student progress and facilitating understanding through discussions [61]. A educator as facilitator was also seen as the person strengthening communication, ethics, leadership, security and software skills [44]. These conditions point to educators as the agents in charge of developing student-centred learning environments [87].
Educators' skills and competencies were also discussed in relation to the development of more specific expertise, including the use of social network analysis techniques to understand social relationships when students are part of an online network or community [12]. Furthermore, several articles referred to the use of a peer learning assistance approach, that is, having peers to hold office hours, help with labs and facilitate student group work, as they were shown to better support learning than teaching assistants [67], and the use of specific educational games [15,87] and remote laboratories [14] that could support CS education. In terms of game-based approaches to CS, educators should have the skills to provide tailored and personalised feedback [87] and assign students to game roles within a course management system [15].

| DISCUSSION
This study aimed to catalogue and synthesise 'innovative' teaching and learning approaches within CS by aligning it to the relatively new concept of Education 4.0 [7,79].
Building on a small-scale initial study conducted among European CS educators [71], this review used a threestep coding process to review and synthesise 66 CS studies from across the globe in the period 2016-2020. In line with our initial study [71], in terms of RQ1-2 it seems that across the globe there are roughly three different design flavours (i.e., EDU4 light, project/hands-on, Full EDU4) with regard to innovative practice in teaching and learning in CS.
While none of the 66 studies explicitly mentioned 'education 4.0', which may be due to its recent conceptualisation, all these studies used some form of Education 4.0 or EDU4 characteristics by Hussin [7]. EDU4 light studies mostly focussed on more independent learning any time/anywhere, personalised learning, and choice of how to learn. As the descriptions of these studies show, a large number of technical and teaching innovations have been introduced into the CS curriculum, although they were mainly focused on one or two EDU4 rather than the full nine features. This may be related to the educator's willingness to make some innovations based on the specific problems found in a course, that is to 'update' part of the teaching method instead of completely redesigning a CS course [53,63].
The second cluster had a strong focus on projectbased learning and hands-on learning. These studies primarily used collaborative and project-based learning approaches with some interesting innovations, such as where CS students were considered as potential entrepreneurs [31]. In all 21 studies, great emphasis was placed on practice and project-based learning, enabling CS graduates to develop strong programming and soft skills, usually working together as a team. However, due to the nature of project-based learning, flexibility in anytime, anywhere, personalisation and learning options were often relatively low.
The third cluster encompassed elements from both the first and second clusters and was mainly centred on hands-on learning, more independent, personalised learning, learning any time/anywhere and choice of how to learn. The lowest characteristic of EDU4 in this cluster was the student's ownership of curriculum, although this is significantly higher than the other two clusters. Several innovative and integrated perspectives used game-based learning [87], flipped classrooms [62] and online lab work [14], showing how CS educators can help students to develop coding, project and team skills.
While the three clusters might suggest clear differences in flavour in terms of the EDU4 characteristics, as is evident from both Tables 2-4 and the narratives of the illustrated studies substantial differences were present within each of the clusters in terms of focus, CS skills, technologies implemented and pedagogical innovations. Despite the substantial differences within the practices in each cluster, perhaps the relatively clear cluster formations might indicate several different stages of development and innovation in a respective context. Indeed, in some learning design research, there is emergent evidence of common design practices [53,58] when comparing different disciplines. In other words, for some contexts, just a small tinkering to a model that mostly works might be appropriate, while in other contexts CS educators might go for a full overhaul of their practice.
In terms of RQ3, about half of the studies made an explicit reference to skills and competencies CS educators should have or develop to align their CS teaching to Education 4.0. Perhaps interestingly, those studies that did refer to the skills and competencies of educators on average had significantly higher scores on nearly all EDU4 characteristics. This might indicate that CS authors who employed more innovative pedagogical approaches, in particular when implementing flipped classrooms or interactive games or lab-exercises, felt the need to provide more detailed narratives about their peer educators, and needed to be aware of the need for additional skills and competences to implement these innovative approaches.
Based upon our systematic literature review, we encourage researchers to investigate how CS courses could be transformed to include more (and perhaps even all) Education 4.0 characteristics, such as those identified by World Economic Forum [93], and whether this would lead to improved skills of CS graduates. It is important for engineering to follow trends in industry, while also providing future anticipation of possible changes. Furthermore, there is an urgent need to critically assess whether the concept of Education 4.0 is useful (or not) for CS and engineering in general. While some of the concepts of active learning, empowering students and hands-on learning are increasingly common and wellsupported by robust evidence, more research is needed to explore whether all elements of Education 4.0 are necessarily beneficial for learning or not.

| Limitations and future research
There is an inherent systemic bias in terms of published outputs, as it is more likely that successful innovations and experiments are published than unsuccessful innovations, as well as 'business as usual' approaches. Furthermore, with the rapid changes in engineering and CS in particular, and the shift in practice due to coronavirus disease 2019 (COVID-19), the reported findings might evolve over time, and perhaps (sub) disciplinary differences might be observed within engineering and CS.
Another limitation is the search string that was used, whereby different key terms of search strings might have resulted in different outcomes. Nonetheless, using a robust 3 phase coding strategy, we believe that we are the first to systematically review the pedagogical learning design decisions that CS educators make when designing innovative practices. Future research should establish which of these common design practices work well for which groups of CS students, and for which specific knowledge, skills and competencies. This will help to strengthen our evidence base and understanding of how to effectively design innovative CS courses that help to empower Education 4.0 in Industry 4.0.

| Implications for teaching and learning
By using the Education 4.0 characteristics our findings suggest three common flavours that CS educators use to design their practice. As evidenced from this study learning practices associated with Education 4.0 require considerable time for preparation compared with the traditional lecture-based class and this may be overwhelming for some teachers, especially those who have not practised Education 4.0 approaches in the past. Therefore, support should be provided through, for example, teaching assistants, fellow teachers or the reuse of existing activities to help teachers to gradually develop the proposed skills and competences. For teachers comfortable with implementing innovative designs adopting some of the applications and approaches developed within studies of the project-based/hands-on cluster or perhaps even the full EDU4 cluster might be appropriate. For teachers less familiar with innovative design, implementing some of the design elements of EDU4 light studies might be useful.