Artificial Intelligence and Learning Analytics in Teacher Education: A Systematic Review

Abstract

In recent years, artificial intelligence (AI) and learning analytics (LA) have been introduced into the field of education, where their use has great potential to enhance the teaching and learning processes. Researchers have focused on applying these technologies to teacher education, as they see the value of technology for educating. Therefore, a systematic review of the literature on AI and LA in teacher education is necessary to understand their impact in the field. Our methodology follows the PRISMA guidelines, and 30 studies related to teacher education were identified. This review analyzes and discusses the several ways in which AI and LA are being integrated in teacher education based on the studies’ goals, participants, data sources, and the tools used to enhance teaching and learning activities. The findings indicate that (a) there is a focus on studying the behaviors, perceptions, and digital competence of pre- and in-service teachers regarding the use of AI and LA in their teaching practices; (b) the main data sources are behavioral data, discourse data, and statistical data; (c) machine learning algorithms are employed in most of the studies; and (d) the ethical clearance is mentioned by few studies. The implications will be valuable for teachers and educational authorities, informing their decisions regarding the effective use of AI and LA technologies to support teacher education.

1. Introduction

Educational research on artificial intelligence (AI) and learning analytics (LA) has been growing in recent years. In particular, AI is having a significant impact in fields such as medicine, finance, and industry [1,2], and education is no exception. Today, research is also focusing on the application of AI and LA technologies in education [3]. Moreover, teacher education has been gradually introducing the use of emerging technologies to train both pre-service teachers (PSTs) and in-service teachers (ISTs). For example, teacher education has been changing from traditional classes—which are no longer the only channel by which students are to be instructed—to include online courses. Furthermore, the widespread adoption of massive open online courses (MOOCs) has made it possible to analyze student engagement based on their activities, which are tracked through the platform using analytics [4]. In the same way, AI techniques, such as natural language processing, have been used to analyze text and oral discourse [5].

1.1. Teacher Education

Teacher education is defined as the practices, strategies, and policies that prepare teachers with the professional knowledge, teaching skills, evaluation techniques, and ethical orientations needed to effectively perform their teaching activities in order to contribute to the development of society [6]. Teacher education is usually considered to have three phases—pre-service, induction, and in-service—all of which are part of a continuous process [7]. Thus, teacher education means both the basic and foundational teacher education oriented towards pre-service teachers and continuous teacher education oriented towards in-service teachers who receive professional development training. Regarding the use of technology, most teachers now recognize the importance of technology in teaching and learning activities. Thus, teacher education programs integrate technology in different ways within the classroom or via online courses—for example, by employing social media, blogs, web conferences, and discussion forums. However, the integration of technology into courses is still difficult due to several factors, such as the school culture, availability of resources, and teachers’ attitudes, knowledge, and skills [8,9]. Nevertheless, governments around the world are implementing policies to bring technology to classrooms, as it is becoming an essential component of the education system [10,11]. Therefore, teacher education plays an important role in developing teachers’ knowledge and skills related to the use of technology in the classroom.

1.2. Artificial Intelligence in Education

AI can be defined as “computing systems that are able to engage in human-like processes such as learning, adapting, synthesizing, self-correction and use of data for complex processing tasks” [12] (p. 1). AI has many branches and sub-branches, such as (a) machine learning (ML), which consists of algorithms that use educational data to identify patterns through successive training with the data; (b) deep learning, which uses large datasets to simulate and predict educational outcomes; and (c) natural language processing (NLP), which employs algorithms for language recognition to extract and analyze textual meaning [13]. In education, AI supports and enhances learning environments by employing intelligent tutoring systems, intelligent agents, and intelligent collaborative learning systems. Recently, the education sector has been significantly influenced by AI research [14], and an interdisciplinary approach is required to integrate several fields, including computer science, image processing, linguistics, psychology, and neuroscience. AI supports teachers’ decision making by reporting real-time class statuses and responding to students’ needs through personalized learning platforms. Moreover, AI has the potential to change the education system [15].

1.3. Learning Analytics

Learning analytics (LA) is defined as the “measurement, collection, analysis and reporting of data about learners and their contexts, for purposes of understanding and optimizing learning and the environments in which it occurs” [16] (p. 4). LA builds upon research areas such as educational data mining, data visualization, recommender systems, and personalized adaptive learning. The foundation of LA is educational data, as the first step is to collect data from various educational environments to identify indicators/metrics. Then, LA techniques are applied to explore and discover useful patterns. This step is followed by monitoring data, and then performing the analysis and prediction; in other cases, the data-driven interventions can lead towards the adaptation and personalization of the learning experiences, and further lead towards reflection. Therefore, the main focus of LA is to capture and analyze students’ actions generated in the learning environment to improve and enhance learning and teaching practices [17]. LA usually has two analytical approaches: (a) descriptive analytics, which is focused on the data-based actions that learners leave behind when they employ digital tools or interact in online platforms, and (b) predictive analytics, which predicts educational outcomes, such as dropout rate, based on students’ behavioral data, historical data (e.g., past course grades), and sociodemographic data [18]. For example, today, specialized educational applications are being designed for personalized learning [19], new communication tools are enabling interaction and professional collaboration between teachers [20], and digital technologies are becoming more embedded across the education sector, influencing the work of teachers [21]. In other words, teachers entering the profession need to be prepared for increasingly digitized education.

It is important to understand that AI and LA models can support teachers, through the provision of educational applications, in the same way as these technologies are reshaping other fields, e.g., medicine. Tondeur et al. [22] have highlighted the need to prepare the next generation of teachers for the integration of technology in education. Moreover, several governments have launched technology policies [23] recommending early awareness for AI. However, how to employ the new technologies—and especially AI—in education is still a gray area [24], and requires teachers to be prepared for the introduction of advanced technologies in education. Thus, teacher education plays an important role in the preparation of teachers for the future [25].

In this context, this study aims to provide an overview of the research on AI and LA in teacher education, with the specific objective of summarizing recent studies in the field, through the identification of their main goals, data sources, techniques and tools employed, the participants, and ethical procedures carried out by the studies. It is important to understand how AI and LA are impacting teacher education, in order to guide teachers, practitioners, and decision-makers regarding the potential of new technologies to support teacher education. As Garbett and Ovens [26] highlighted, teacher education needs not only to focus on pedagogical knowledge to function in the schools, but also to equip pre-service teachers to operate in an increasingly digital world.

This study applies a systematic review methodology to explore the relevant literature on the use of AI and LA technologies to improve teacher education, and it is organized as follows: First, it presents the introduction, followed by the research methodology and the review’s results. Then, we discuss the findings and implications for the future of teacher education. Finally, the conclusions are presented. The research questions that guide the present study are as follows:

• RQ1: What are the main goals and objectives of the reviewed studies regarding the use of AI and LA in teacher education?
• RQ2: What kinds of data sources are employed by the studies on AI and LA in teacher education?
• RQ3: What kinds of AI and LA techniques and tools are used to support teacher education?
• RQ4: Who are the participants included in the studies on AI and LA in teacher education?
• RQ5: How are ethical procedures being fulfilled by studies on AI and LA in teacher education?

2. Methodology

This systematic literature review was carried out to provide up-to-date information on AI and LA in teacher education. A systematic review is an explicit and systematic process for identifying, extracting, and synthesizing knowledge gained from a variety of empirical studies to answer research questions [27]. Moreover, Sleeter [28] highlighted the need to carry out systematic reviews on teacher education to provide a more comprehensive understanding of questions that remain under-researched.

The present systematic review follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [29]. The following databases were searched because they cover a broad range of educational journals: Web of Science, which is a major research platform that provides access to leading academic literature; ScienceDirect, which is a comprehensive collection of scientific journals; and IEEE Xplore, which is a large database of scientific and technical articles. The search strings included the following terms: (“teacher education” OR “pre-service” OR preservice OR “in-service”) AND (“learning analytics” OR “artificial intelligence” OR “machine learning” OR “deep learning”). They were input into the applications of the aforementioned databases and used as a filter for documents that included the search strings in their title, abstract, and/or keywords. In line with the PRISMA guidelines, the following criteria were used to decide which articles to include in the final revision: (a) articles published from 2017 to 2021; (b) articles published in the English language; (c) articles presenting empirical, primary research; (d) articles involving pre-service teachers (PSTs) or in-service teachers (ISTs); and (e) articles exclusively related to AI or LA in teacher education (

Table 1. Inclusion and exclusion criteria.

Inclusion Criteria	Exclusion Criteria
Published from 2017 to 2021 English language Empirical, primary research Research involving pre-service teachers (PSTs) or in-service teachers (ISTs) in teacher education programs Studies related to AI or LA in teacher education (AI with a focus on machine learning, deep learning, and natural language processing applications)	Published prior to 2017 Not in English Not empirical; not primary research (e.g., reviews) Research not involving pre-service teachers or in-service teachers in teacher education programs Studies not relating to LA or AI in teacher education

).

The search generated 2012 articles, whose titles were screened, after which 325 potential articles remained. Another 27 duplicates were eliminated, leaving 298 articles for abstract screening. Then, after applying the inclusion and exclusion criteria, 73 articles were eligible for full screening; however, 43 of these were irrelevant to the topic of this study. Finally, 30 articles remained to be analyzed (Figure 1).

Data were extracted by collecting and coding the information for each of the 30 selected studies and identifying the following information to help organize the data for analysis: author(s), publication year, goals and objectives of the study, data sources employed and their characteristics, techniques and digital tools employed in the study, sample size and type of participants, whether the study obtained the informed consent of the participants, and the results of the study. The selected papers were analyzed through inductive and deductive processes, including reading and re-reading through data to identify themes for analysis [30]. Two researchers were involved in the procedures, and reviewed each article independently, achieving inter-rater reliability of 81% (Cohen’s kappa coefficient). Disagreements were resolved by discussion until agreement was reached. The emerged themes are displayed using frequency tables.

Table 2. Summary of the studies included in this review.

Author(s) and Year	Country/Region	Goals and Objectives	Participants	Data Sources	Techniques	Tools	Ethical Procedures	Results
Bao et al. (2021) [31]	China	To visualize students’ behaviors and interactions.	35 PSTs	Knowledge elaboration (K): discussion of topics and posts Behavior patterns (B): posting frequency; posts’ content Social interaction (S): network density, network cohesion, and network interactions	LA dashboard	Moodle platform LA dashboard Knowledge–Behavior–Social Dashboard tool (KBSD)	n.d.	The KBSD tool has the potential to assist teachers in detecting learning problems. The most common strategy was cross-group; the interventions involved cognitive guidance, scaffold instruction, and positive evaluation.
Benaoui and Kassimi (2021) [32]	Morocco	Perceptions of PSTs’ digital competence.	291 PSTs	DigComp framework (five dimensions): information and data literacy, communication and collaboration, digital content creation, safety, and problem solving	AI machine learning	AI tools: K-means clustering	n.d.	PSTs felt competent when using digital technologies daily, but they did not feel competent in digital content creation and problem solving. This might be due to the predominance of theoretical knowledge at the expense of real practice in teaching training.
Chen (2020) [33]	China	To investigate whether visual learning analytics (VLA) has a significant influence on teachers’ beliefs and self-efficacy when guiding classroom discussions.	46 ISTs	Video discourse data: number of words; number of turns; teacher–student turn-taking patterns Visualizing talk strategies: elaborating, reasoning, listening, and thinking with others	LA visual learning analytics (VLA)	Classroom discourse analyzer (CDA) is a VLA tool that automatically extracts and visualizes low-inference discourse information	n.d.	The VLA approach to video-based teacher professional development had significant effects on teachers’ beliefs and self-efficacy, and influenced their actual classroom teaching behavior.
Cutumisu and Guo (2019) [34]	Canada	To determine PSTs’ knowledge of and attitudes toward computational thinking through the automatic scoring of short essays.	139 PSTs	Short essays (500-word reflection) about the experience of solving a block-based visual programming scenario were analyzed	AI machine learning	Moodle platform Latent Dirichlet Allocation library in Python. Code.org	PSTs provided informed consent	Topics that emerged from PSTs’ reflection included assignment (66.7%), skill (11.6%), activity 10.1%), and course (6.5%).
Fan et al. (2021) [35]	China	To reveal links between learning design and self-regulated learning.	7030 PSTs, 1758 ISTs	Number of sessions, duration, number of actions, etc. Content access, content revision, discussion, assessment, help-seeking, and search	LA	MOOC R package AI tools: Expectation-maximization (EM) algorithm; Bayesian information criterion (BIC); TraMineR	n.d.	Four meaningful learning tactics were detected with the EM algorithm: search (lectures), content and assessment (case-based or problem-based), content (project-based), and assessment.
Hayward et al. (2020) [36]	Canada	To explore PSTs’ engagement with models of universal design for learning and blended learning concepts.	197 PSTs	Access features: location, date, time, and regularity (average number of logins/week) Content features: screencasts; quizzes	LA	Moodle platform	n.d.	The feature regularity of access had a moderate relationship with student engagement. High achievers tended to have a set of strategies.
Hsiao et al. (2019) [37]	China Taiwan	To assess the qualities of pre-service principals’ video-based oral presentations through automatic scoring.	200 pre-service principals	Video-based speech features: content, speech organization, appropriate word usage, proper etiquette, correct enunciation, fluent prosody, timing control	AI machine learning	Supervised algorithms: support-vector machine (SVM) classifier, logistic regression, random forests, and gradient-boosted decision trees	n.d.	The SVM classifier had the best accuracy (55%). It was found that human experts can potentially suffer undesirable variabilities over time, while automatic scoring remains robust and reliable over time.
Ishizuka and Pellerin (2020) [38]	Japan	To assess real-time activities in second language classrooms.	4 PSTs	Class, group, and individual work Student modalities: reading, writing, listening, and speaking Material: extended, minimal, native, or non-native	AI	Video on the ePortfolio in Moodle AI mobile communicative orientation of language teaching (COLT) scheme	n.d.	The integration of AI mobile COLT analysis has strong potential to follow-up PSTs’ progress throughout their practicum.
Jensen et al. (2020) [39]	USA	To provide automated feedback on teacher discourse to enhance teacher learning.	16 ISTs	Discourse data: recordings of classroom conversations. Included variables: specificity, instructional talk, authentic questions, dialogic, cognitive level	AI machine learning	Random forests (RF) classifier and regression IBM Watson AI speech recognizer	n.d.	The RF classifier had 89% accuracy, generating automatic measurement and feedback of teacher discourse using self-recorded audio data from classrooms.
Karunaratne and Byungura (2017) [40]	Rwanda	To track in-service teachers’ behavior in an online course of professional development.	61 ISTs	User action data: time, full names, event context, components, event names, activity, IP address and origin Performance data: grades	LA visual learning analytics	Moodle platform R software	n.d.	Half of the registered teachers never accessed the course. Most of the teachers were actively engaging in the virtual learning environment’s activities.
Kasepalu et al. (2021) [41]	Estonia	Teachers’ perceptions of collaborative analytics using a dashboard based on audio and digital trace data.	21 ISTs	Trustworthiness (0–100), novelty, and usefulness Actionability and receiving new information Level of experience	LA dashboard	CoTrack: a Raspberry-Pi-based prototype with microphones CoTrack’s dashboard showing speaking time and social networks Etherpad	Consent forms were filled out by ISTs and their students	New information enhances teachers’ awareness, but it seems that the dashboard decreases teachers’ actionability. Therefore, a guiding dashboard could possibly help less experienced teachers with data-informed assessment.
Kelleci and Aksoy (2020) [42]	Turkey	To examine PSTs’ and ISTs’ experiences using an AI-based-simulated virtual classroom.	16 PSTs, 2 ISTs	Discourse data and reflection elements: PSTs’ attitudes, experiences, device preferences, comments about the interface, and content and technical issues	AI simulation	SimInClass: an AI-based-simulated virtual classroom Google Classroom learning platform	Ethical approval from the institution	The SimInClass simulation was effective in providing clear directions and giving feedback. PSTs suggested that the simulation should give clues as to correct solutions.
Kilian et al. (2020) [43]	Germany	To predict PSTs’ dropout for a mathematics course and identify risk groups.	163 PSTs	Performance: GPA, math grade, TIMSS, age, gender, federal state, school type, type of student	AI machine learning	AI tools: SVM, LR, LR with elastic net regularization, and tree-based methods	PSTs provided written informed consent	Risk level 1: score ≤ 12 (highest risk), GPA > 2.1; risk level 2: score ≤ 12 (high risk), GPA ≤ 2.1; risk level 3: score > 12 (moderate), 1.6 < GPA ≤ 2.
Kosko et al. (2021) [44]	USA	To examine PSTs’ professional noticing of students through video and ML.	6 PSTs, subsample of 70 PSTs	Behavior patterns: recordings of PSTs’ viewing a 360 degrees video with students’ actions Short writings: PSTs select one pivotal moment and explain why it is significant	AI machinelearning	AI tools: machine learning algorithm	n.d.	PSTs’ actions relevant to pedagogical content-specific noticing could be detected by AI algorithms. PSTs’ behavior may have been due to professional knowledge rather than experience.
Lucas et al. (2021) [45]	Portugal	To measure teachers’ digital competence and its relation to personal and contextual factors.	1071 ISTs	Digital competence areas: Personal: age, gender, teaching experience, confidence, and years using digital technology in teaching Contextual: classroom equipment, students’ access to technology, network infrastructure, and curriculum Professional engagement: digital resources, assessment, empowering learners, and facilitating learners’ digital competence	AI machine learning	SPSS STATA, fast-and-frugal trees (FFTrees) classifier in machine learning	(Voluntary and anonymous teachers)	For personal factors, FFTrees had an accuracy of 81%, while for contextual factors it was 66%. For digital competence, the important personal factors were the number of digital tools used, ease of use, confidence, and openness to new technology. The contextual factors included students’ access to technology, the curriculum, and classroom equipment.
Michos and Hernández-Leo (2018) [46]	Spain	To support community awareness to facilitate teachers’ learning design process using a dashboard with data visualizations.	23 PSTs, 209 ISTs	ILDE dashboard: profile views, comments, created designs, re-used designs, and edits.	LA dashboard	The Integrated Learning Design Environment (ILDE) dashboard IBM SPSS 22 Heidi SQL and Tableau	n.d.	The ILDE dashboard can provide an understanding of the social presence in the community of teachers. Visualization was the most commonly used feature. There were time constraints.
Montgomery et al. (2019) [47]	Canada	To examine the relationships between self-regulated learning behaviors and academic achievements.	157 PSTs	Self-regulated behaviors: Activating: online access location, day of the week, time of day Sustaining: access frequency Structuring: average logins per week, exam review patterns, number of reviewed quizzes/day	LA	Moodle platform	n.d.	84.5% of PTSs’ access to the platform took place off-campus. The strongest predictors for student success were the access day of the week and access frequency.
Newmann et al. (2021) [48]	Germany	To support PSTs’ self-study using chatbots as a tool to scale mentoring processes.	19 PSTs	Social bot: user intentions, bot messages System Usability Scale: frequency, ease of use, confidence, consistency	AI NLP	Chatbots: Feedbot for self- study, Litbot for mentoring students’ reading	n.d.	Promising results that bear the potential for digital mentoring to support students.
Post (2019) [49]	USA	To challenge PSTs to analyze and interpret data on students’ online behavior and learning.	n.d. PSTs	Learning action logs about search terms, visited websites, time spent on each website, and the order in which sites were visited	LA	Thinking app (Chrome extension) that tracks online behaviors	n.d.	PSTs lacked media literacy skills. Online assignments promoted student-centered learning and critical thinking. The prevalence of multitasking was highlighted.
Pu et al. (2021) [50]	Malaysia	To design a service-learning-based module training AI subjects (SLBM-TAIS).	60 PSTs	Psychological variables: Practical knowledge: educational beliefs, interpersonal relationships, teaching strategies, self-reflection Motivation: intrinsic motivation, extrinsic motivation, amotivation Other: gender, teaching experience, average academic performance	AI	The SLBM-TAIS educational module	n.d.	The SLBM-TAIS was effective in training PSTs to teach AI subjects to primary school students. The SLBM-TAIS module influences situational knowledge, teaching strategies, and both intrinsic and extrinsic motivation.
Sasmoko et al. (2019) [51]	Indonesia	To determine teacher engagement using artificial neural networks.	10,642 ISTs	Based on the Indonesian Teacher Engagement Index (ITEI): positive psychology, positive education, teacher performance, nationalistic character, and leadership engagement	AI machine learning (ANN)	Django: a website framework for Python Chart.js for data visualization. MongoDB as the database	Not applicable	The ANN classification accuracy was 97.65%, proving the reliability of the instruments and websites; however, this still requires further testing in terms of both ease of use and trials with diverse data.
Sun et al. (2019) [52]	China	To investigate changes in PSTs’ concept of engagement, analyzing data recorded during PSTs’ discussions via an MOOC platform.	53 PSTs	Discussion data Dimensions based on Bloom’s taxonomy: remember, understand, apply, analyze, evaluate, create	LA	MOOC platform	n.d.	The most frequent discussion behaviors were evaluated (31.52%) and analyzed (27.77%). PSTs with an analytical style implemented multiple strategies for learning.
Vazhayil et al. (2019) [53]	India	To introduce AI literacy and AI thinking to in-service secondary school teachers.	34 ISTs	Types of AI tasks: text recognition, sentiment analysis, image classification, categorical/numerical data	AI	IBM Watson AI model Mitsuku chatbot Google AI experiment named Emoji Scavenger Hunt Scratch	15 ISTs consented to recorded video testimonials	77% appreciated peer teaching, 41% preferred the game-based approach, and 24% were concerned about internet access. The best strategy was embracing creative freedom and peer teaching to boost learners’ confidence.
Wulff et al. (2020) [54]	Germany	To employ AI algorithms for classifying written reflections according to a reflection-supporting model.	17 PSTs	Reflection elements: circumstances, description, evaluation, alternatives, consequences	AI natural language processing	Doc2Vec features Four classifiers: decision trees, multinomial logistic regression, multinomial naïve Bayes, stochastic gradient descent	PSTs provided informed consent	The multinomial logistic regression was the most suitable classifier (0.63). Imprecise writing was a barrier to accurate computer-based classification.
Yang et al. (2020) [55]	China	To enhance self-directed reflective assessment (SDRA) using LA.	47 PSTs	Epistemic agency, democratic knowledge, improvable ideas, reflective and transformative assessment, and community knowledge	LA	Knowledge Forum (online notes)	Ethical approval was obtained from the hosting institution	SDRA fostered PSTs’ collective empowerment, as reflected by their collective decision making, synthesis of ideas, and “rising above” ideas.
Yilmaz and Yilmaz (2020) [56]	Turkey	To examine PSTs’ perceptions of personalized recommendations and feedback based on LA.	40 PSTs	LMS log data: date, login frequency, views per week, participation in discussions	LA	Moodle LMS platform	(Voluntary participation)	LA helped to identify learning deficiencies, provided self-assessment and personalized learning, improved academic performance, and instilled a positive attitude toward the course.
Yoo and Rho (2020) [57]	Korea	To determine ISTs’ training and professional development using ML.	2933 ISTs, 177 principals	Based on the Teaching and Learning International Survey (TALIS) 2013: types of activities, participation rates, intensity of participation, mentoring and induction programs	AI machine learning	Group Mnet technique (glmnet package). R software	Not applicable	Identified 18 predictors of ISTs’ professional development. Found 11 new predictors related to ISTs’ pedagogical preparedness, feedback, and participation.
Zhang J. et al. (2021) [58]	China	To build an intelligent assessment system of PSTs teaching competency.	240 PSTs	PSTs’ teaching competency framework (six dimensions): professional foundation, instructional design, teaching implementation, technology application, teaching evaluation, reflective development	AI machine learning	AI tools: Back Propagation (BP) neural network Delphi and Analytic Hierarchy Process (AHP) methods Matlab software	n.d.	The trained model can be used to evaluate PSTs’ competency on a large scale, its relative error was small between 0–0.2.
Zhang S. et al. (2021) [59]	China	To automatically detect the discourse characteristics of in-service teachers from online textual data.	1834 ISTs	Discourse characteristics: number of posts per teacher, length of post per teacher, much or little new information, high or low topic relevance	AI natural language processing	Word2vec toolkit to generate lexical vectors based on AI-NLP	Ethical approval from the institution	New and relevant information was posted at the beginning of the online discourse. Cluster analysis showed three different posts: relevant topic with new information, another with little new information, and a less relevant topic with little new information.
Zhao et al. (2021) [60]	China	To improve the outdoor learning experience and build a learning resource based on ontology information retrieval.	38 PSTs	Vision-based mobile augmented reality from the university campus (e.g., plants, flowers, trees) through scene detection, retrieval, superposition, visualization, and interaction	AI vision-based mobile augmented reality (VMAR)	MobileNetV2 network: a lightweight convolutional neural network by Google for mobile devices	n.d.	PSTs perceived the usability as good; it was preferred by younger users, and had a positive impact on learning. The average precision of retrieval based on keywords (97.46%) and ontology (90.85%) signified good performance.

Note. CDA = classroom discourse analyzer, DT = decision tree, FFTrees = fast-and-frugal trees, GBTD = gradient-boosted decision trees, KBSD = Knowledge–Behavior–Social Dashboard, ITEI = Indonesian Teacher Engagement Index, RF = random forests, NLP = natural language processing, SLBM-TAIS = service-learning-based module training AI subjects, SVM = support-vector machine, WISE = web-based inquiry science environment.

presents a summary of the analysis carried out with the selected studies.

3. Results

This review includes 30 studies based in 16 countries/regions, with the following distribution: Canada (3), China (8), Estonia (1), Germany (3), India (1), Indonesia (1), Japan (1), Korea (1), Malaysia (1), Morocco (1), Portugal (1), Rwanda (1), Spain (1) China Taiwan (1), Turkey (2), USA (3). The analysis guided by the research questions provides some insights into the impact of AI and LA on teacher education.

3.1. Goals and Objectives

RQ1: What are the main goals and objectives of the reviewed studies regarding the use of AI and LA in teacher education?

Several goals and objectives are mentioned in the selected studies, and can be categorized under six themes: (a) behavior when using AI and LA, (b) digital competence, (c) perceptions of AI and LA, (d) self-regulation and reflection, (e) engagement, and (f) analysis of educational data (

Table 3. Goals and objectives across reviewed studies.

Goals and Objectives	Reference Number	Number of Studies
Behavior when using AI and LA	[31,35,39,40,42,44,52]	7
Digital competence	[34,38,45,50,53,58]	6
Perception of AI and LA	[32,41,48,56,60]	5
Self-regulation and reflection	[33,35,47,54,55]	5
Engagement	[36,51,57,59]	4
Analysis of educational data	[37,43,49]	3

).

The findings indicate that the most prominent category regarding the goals and objectives of the selected studies is the pre- and in-service teachers’ behavior when using AI and LA. For example, some of the studies include the visualization of the behaviors and interactions [31,40,52], where AI and LA tools help to explore the effectiveness of the learning activities offered. This is important, because the behavior of pre- and in-service teachers has received an increased focus, due to its relation to teacher preparation and professional training [61]. Similarly, Tezci [62] highlights how pre- and in-service teachers’ behaviors with regard to new technologies are related to their intentions to use those technologies in the classroom. The next most common category is digital competence, which is generally defined as a set of knowledge, skills, and attitudes required when using new technologies to create, communicate, and resolve problems in an efficient and effective way, and also to improve the teaching process when using technology [63]. For example, in our findings, some of the studies are focused on measuring the digital competence [37,45,53]. Similarly to these findings, a review carried out by Wilson, Ritzhaupt, and Cheng [64] found that pre- and in-service teachers’ digital competence should be considered a necessary skill in their teaching activities. Another category that emerged from the analysis is the perception of AI and LA. For example, some studies examined pre- and in-service teachers’ perceptions about the use of LA [56], as well as their perceived usability of AI-based outdoor learning tools [60]. Cooper et al. [65] also confirmed this conclusion, indicating that pre-service teachers’ positive perceptions of new technologies are related to the potential of technology to enhance learning experiences that they might otherwise not experience with other learning tools. Additionally, the studies mentioned other research goals and objectives, such as self-regulation and reflection, engagement, and the analysis of educational data. For example, some studies focused on the influence of LA feedback on reflection [33,54,55], or on exploring the levels of engagement using LA techniques [36,59] or AI methods [51]. Meanwhile, other reviewed studies even challenged pre-service teachers in order to analyze and interpret educational data [49]. These findings are similar to those of Reeves and Chiang [66], who highlighted that, in recent years, educational data have been used by teachers to inform their practices.

3.2. Data Sources

RQ2: What kinds of data sources are employed by the studies on AI and LA in teacher education?

Data are the foundation of AI and LA, as both require large datasets to perform analyses. Therefore, it is necessary to know the source of the educational data used by the reviewed studies. Three sources of data were identified: (a) behavioral data, (b) discourse data, and (c) statistical data. Each source of data includes different types of datasets (

Table 4. Data sources of the reviewed studies.

Data Sources	Types of Datasets	Reference Number
Behavioral data	Access data	[32,35,36,38,40,46,47,48,49,56,60]
	Social interaction data	[31,44,56]
Discourse data	Text discourse data	[34,53,54,55,59]
	Audio video discourse data	[32,37,39,44]
	Discussion data	[31,42,52]
Statistical data	Sociodemographic data	[35,43,45,50,51,57,58]

). Some of the studies had more than one type of dataset.

Most of the studies employed behavioral data, which were collected by observing and recording pre- and in-service teachers’ behaviors. The selected studies employed two types of datasets: (i) access data, e.g., location, date, time, regularity (i.e., average number of logins per week), number of times quizzes were reviewed per day; and (ii) social interaction data, e.g., network density, network cohesion, and network interaction. These kinds of data help to identify frequent access patterns, and can be properly analyzed. Discourse data are another data source found in the reviewed studies, including three types of discourse datasets: (i) text discourse data, e.g., number of posts per teacher, length of post per teacher, much or little new information, and high or low topic relevance; (ii) audio video discourse data, e.g., number of words, number of turns, and teacher–student turn-taking patterns; and (iii) discussion data, e.g., discussion topics (i.e., time, key terms, frequency), number of posts, posting frequency, and posts’ content. The other type of data source is statistical data; for example, the selected studies presented sociodemographic data related to age, gender, teacher experience, work status, number of tools used in the classroom, and number of years spent using digital technology in teaching. These findings coincide with the results of an existing study carried out by Zhao et al. [67], who noted that digital learning platforms can store massive amounts of learners’ behavioral data, which could help to assess learning and predict learners’ performance.

3.3. Techniques and Tools

RQ3: What kinds of AI and LA techniques and tools are used to support teacher education?

We found that 17 studies mainly used AI techniques (

Table 5. AI and LA techniques across the reviewed studies.

Techniques	Reference Number	Number of Studies	Total
Artificial Intelligence
Machine learning	[32,34,37,39,44,45,51,57,58]	9	17
Natural language processing (NLP)	[48,55,59]	3
Vision-based mobile augmented reality	[38,60]	2
Other AI techniques (e.g., IBM Watson AI)	[42,50,53]	3
Learning Analytics
Dashboards	[31,40,46]	3	13
Visual learning analytics (VLA)	[33,40]	2
Other LA techniques (e.g., apps)	[35,36,47,49,50,52,54,56]	8

); 9 of these used machine learning, 3 employed natural language processing (NLP), 2 used vision-based mobile augmented reality (VMAR), and 3 used other AI techniques. For example, the studies used AI techniques to automatically score video presentations [37], to identify at-risk students and predict the number of dropouts [43], or to classify written reflections [54]. Moreover, 13 studies used LA techniques: 3 studies used dashboards, 2 used visual learning analytics (VLA), and 8 employed other LA techniques. For instance, LA techniques helped to visualize pre-service teachers’ behaviors and interactions [31], to support community awareness and social presence [46], and to investigate changes in engagement [52]. Similar conclusions were mentioned by Verma, Kumar, and Kohli [68], who indicated that AI techniques could benefit and enhance the quality of education.

Additionally,

Table 6. AI and LA tools across the reviewed studies.

Themes	Tools	Reference Number
AI tools	Algorithms: SVM, LR, NLP, etc.	[32,35,37,39,43,44,45,48,55,57,58,59,60]
	AI systems: IBM Watson AI	[38,39,42,53]
Online platforms	Moodle	[31,34,36,38,40,47,56]
	MOOC	[35,52]
Data analysis tools	Programming language: R, Python	[34,35,40,51,57]
	Statistical software: SPSS, STATA	[45,46,58]
Monitoring tools	Dashboards: WISE, ILDE, KBSD	[31,41,46]
	Modules: CDA, SLBM-TAIS	[33,49,50,54]

Note. CDA = classroom discourse analyzer, KBSD = Knowledge–Behavior–Social Dashboard, LR = logistic regression, NLP = natural language processing, SLBM-TAIS = service-learning-based module training AI subjects, SVM = support-vector machine, WISE = web-based inquiry science environment footer.

lists the AI and LA software tools used in the studies. The most frequently used software tools were AI algorithms, followed by online platforms (such as the Moodle platform) and MOOC courses. Moreover, some of the reviewed studies reported the use of data analysis tools such as programming languages and statistical software, while others used monitoring tools, such as dashboards and modules. For example, these software tools were employed to support pre-service teachers’ self-study using chatbots [48], to determine teachers’ professional development [57], to automatically detect discourse characteristics [59], etc. Namoun and Alshanqiti [69] also found that AI and LA tools could help to visualize and predict learners’ achievements. Moreover, LA tools are becoming common in the context of online learning and blended learning (e.g., MOOCs).

3.4. Participants in the Studies

RQ4: Who are the participants included in the studies on AI and LA in teacher education?

Regarding the participants involved in the reviewed studies, we found that 18 studies had pre-service teacher participants, while 9 studies included in-service teachers who were engaged in teacher education programs, and 3 studies included both pre- and in-service teachers (

Table 7. Participants across the reviewed studies.

Participants	Reference Number	Number of Studies
Pre-service teachers (PSTs)	[31,32,34,36,37,38,43,44,47,48,49,50,52,54,55,56,58,60]	18
In-service teachers (ISTs)	[33,39,40,41,45,51,53,57,59]	9
Both pre- and in-service teachers	[35,42,46]	3

). These findings indicate that not only pre-service teachers, but also in-service teachers, are being taught to use advanced technologies such as AI and LA to update their knowledge, practices, and digital competence, all of which eventually benefit students. Coincidentally, Seufert, Guggemos, and Sailer [70] mentioned that for pre- and in-service teachers, it is important to receive continuous professional development in technologies, so as to enhance their knowledge and skills.

3.5. Ethical Procedures

RQ5: How are ethical procedures being fulfilled by studies on AI and LA in teacher education?

Ethics represent an important issue regarding the use of technology. We reviewed how the selected studies on AI and LA in teacher education fulfilled the ethical procedures when collecting data from pre- and in-service teachers.

Table 8. Data sources of the reviewed studies.

Ethical Consent Granted by	Reference Number	Number of Studies
Pre- and in-service teachers	[34,41,43,53,54]	5
The higher education institution	[42,55,59]	3
Short reference	[45,56]	2
Not applicable	[51,57]	2
Not mentioned	[31,32,33,35,36,37,38,39,40,44,46,47,48,49,50,52,58,60]	18

shows that 5 studies obtained ethical consent from the pre- and in-service teachers themselves, 3 studies were granted ethical consent by the institution where the research was performed, 2 studies made a short reference to the pre-service teachers’ voluntary participation, and 2 studies indicated that ethical procedures were not applicable (e.g., secondary data sources). However, most of the studies (18 studies in total) did not mention ethical procedures. Similar to our findings, other researchers such as Krutka et al. [71] have noted the same issue. Moreover, Stahl et al. [72] also highlighted the need for responsible research and innovation, with an emphasis on data privacy and security.

4. Discussion

Overall, this study provides important insights regarding the status of research on AI and LA in teacher education, which can be summarized as follows:

First, teacher education is constantly adapting and gradually introducing the use of new technologies to both pre- and in-service teachers. The application of digital technologies in education presents both opportunities and challenges. Researchers have mentioned that AI has brought some opportunities for teachers, including automated grading that provides support to lessen teachers’ workload [73], predictive analytics to detect students at risk of not completing a course [74], adaptive learning that identifies areas to provide more focused learning experiences [75], and chatbots that are helpful virtual assistants for teachers [76]. However, research is also highlighting some ethical concerns, such as privacy when compromising the exploitation of data via recommender systems [77], tracking systems that gather detailed information about actions and preferences [78], and bias and discrimination, e.g., perpetuating gender bias and social discrimination [79]. As Cadwell [80] pointed out, technologies may offer several benefits within teacher education, but it is also necessary to emphasize the importance of preparing pre-service teachers to integrate technologies into education. A complex world brings new conditions, where unexpected changes might require pre- and in-service teachers to deliver instruction through the use of technology. Moreover, regarding the use of technology in teacher education, Carrier and Nye [81] highlighted how professional development in the use of technology can empower educators, as it enhances their teaching and supports students’ learning experience [82].

Second, our review presents examples where AI and LA techniques have the potential to assist teachers in several teaching activities. For instance, we found that AI and LA methods can help to visualize PSTs’ behaviors and interactions [31], to predict PSTs’ dropout and identify risk groups [43], to support PSTs’ self-study using chatbots as a tool to scale mentoring processes [48], to automatically detect discourse characteristics from online textual data [59], to assess—through automatic scoring—the qualities of video-based oral presentations [37], to asses PSTs’ teaching competency through an intelligent assessment system [58], to classify written reflections according to a reflection-supporting model [54], etc. Similarly, Goksel and Bozkurt [83] considered that AI-featuring technologies could contribute to the advancement of some educational processes. Regarding LA methods, Van Leeuwen, Teasley, and Wise [84] also arrived at the conclusion that learning analytics can play a constructive role that can enhance and complement teachers’ decision making. However, it is necessary to develop pre- and in-service teachers’ digital competence, as an essential requirement for using advanced technologies in teaching education. Furthermore, Luckin et al. [85] indicated that teachers need to be empowered through adequate training in order to be AI-ready, which means to know how AI could be used to enhance their human teaching capabilities and expertise.

Third, our findings indicate that studies on teacher education—especially when employing AI and LA techniques and tools—need to pay attention to the importance of obtaining consent from the participants. Similarly, Pusey [86] found that pre-service teachers do not possess adequate knowledge or the ability to keep their future students’ data safe from exposure and harm. In other words, ethical issues bring some concerns about cybersecurity in education. Therefore, Reidenberg and Schaub [87], as along with other researchers, have proposed the need for transparency [88], accountability [89], and fairness [90] in the use of AI and LA in education [91]. As Siemens et al. [92] highlighted, ethics and data privacy need careful consideration in LA research. Furthermore, Holmes et al. [91] pointed out the importance of carrying out responsible research on AI in education, and the need for researchers to be trained to tackle emerging ethical questions. Therefore, researchers, students, teachers, and educational authorities should be aware of the importance of ethics with respect to personal data.

This systematic review has some limitations. First, it is focused on teacher education, including pre- and in-service teachers in education programs. Future studies could enhance the scope and include teachers who are not enrolled in education programs. Second, the present review focused on learning analytics and artificial intelligence, mainly including applications of machine learning and deep learning in education. Future studies could include other AI techniques and tools. Finally, it cannot be guaranteed that every relevant article was found; nonetheless, the present study contributes to the analysis of the use of AI and LA in teacher education.

5. Conclusions

This systematic review highlights how AI and LA are being employed in teacher education, as AI and LA techniques are gradually being adopted to support teaching activities at different educational levels. However, the rate of adoption of AI and LA in education is still slow compared to other fields, such as medicine, industry, and finance. The present study provides some evidence-based educational innovations through the application of AI and LA technology in teacher education. These applications have several purposes—for example, to visualize pre- and in-service teachers’ behaviors and interactions, to assess their video-based oral presentations through automatic scoring, to introduce AI literacy to in-service teachers, etc. One issue that is highlighted by this review is the lack of attention to ethics and data privacy, as few of the reviewed studies mentioned ethical clearance. Lastly, it is important that more teachers, practitioners, educational authorities, and decision-makers become involved and understand the opportunities and challenges that AI and LA technologies could bring to teacher education.