1

Journal of Advanced Technology and Multidiscipline (JATM)

Vol. 01, No. 01, Year, pp. 1-1x

Student’s Behavior Clustering based on

Ubiquitous Learning Log Data using
Unsupervised Machine Learning
Ika Qutsiati Utami1, Wu-Yuin Hwang2, Ratih Ardiati Ningrum3
1,3
Data Science Technology, Engineering Department, Faculty of Advanced Technology and Multidiscipline,
Universitas Airlangga, Surabaya, Indonesia
1,2
Graduate Institute of Network Learning Technology, National Central University, Taoyuan, Taiwan
2
Department of Computer Science and Information Engineering, National Dong Hwa University, Hualien, Taiwan

Abstract—Online learning is the source of data generation through instrumental investigation. According to the first
related to learner’s learning behaviors, which is valuable for definition of LA, it is an approach to collecting, analyzing, and
knowledge discovery. Existing research emphasized more on an reporting educational data related to learning, learners, and its
understanding of student’s performance and achievement from related context [7]. There are several techniques used for the
learning log data. In this study, we presented data-driven learning
analytical process of learning related data such as supervised
behavior clustering in authentic learning context to understand
students’ behavior while participating in the learning process. The
and unsupervised learning methods. The main difference
objective of the study is to distinguish students according to their between the two approaches is the use of labeled data (Nafis
learning behavior characteristics and identify clusters of students and Biswas, 2022; Shakarami, Shahidinejad and Ghobaei-
at risk of unsuccessful learning achievement. Learning log data Arani, 2021). In the unsupervised learning method, the process
were collected from ubiquitous learning applications before of data analysis to learn the patterns from data does not require
conducting Exploratory Data Analysis (EDA) and cluster analysis. labeled input and output data. It is being used generally for
We used partitional clustering using K-means algorithm and clustering and segmentation-related tasks. The algorithm
hierarchical clustering based on the agglomerative method to performs natural clustering over the dataset to identify similar
improve clustering strategies. The result of this study revealed
patterns and characteristics. The process of learning about user
three different clusters of students supported by data visualization
techniques. Cluster 1 comprised more students with active
behavior from log data typically involves partitioning the data
learning behavior based on the total logs, total problems posed, into meaningful subsets, called partitions, and comparing the
and the total attempts in fraction operation and simplification. different partitions.
Students in clusters 2 and 3 had a higher attempt at problem- In an educational context, cluster analysis can be used to
solving instead of problem-posing. Both clusters also focused on gain insight into structured data such as student behavior
fraction’s conceptual understanding. Knowledge discovery of this grouping, finding similar learning patterns, and student
study used real data generated from ubiquitous learning performance clustering [10], [11]. However, despite the
application namely U-Fraction. We combined two different types potential of unsupervised learning or cluster analysis for LA, it
of clustering method for delivering more accurate portrait of a
is seldom utilized for supporting teaching and learning analysis
student’s hidden learning behaviors. The outcome of this study can
be a basis for educational stakeholders to provide preventive
in ubiquitous learning contexts based on students’ learning log
learning strategies tailored to a different cluster of students. data [12]. Log data is automatically produced files and
timestamps relevant to the system or software application [13].
Keywords—Learning analytics, behavior clustering, Log data can provide a portrait of a student’s hidden learning
unsupervised learning, learning log-data, education research, behavior and give a more complete or accurate picture of all
educational policies. behaviors. Yet, log data generated by the learning application
server had left the characteristic prone to data noise. The
I. INTRODUCTION process of mining and reducing noise in log data is considered
Over ten years, starting from 2011, learning analytics (LA) as challenging task. In addition to that fact, this study tries to
with data-driven analysis has arisen by exploiting machine perform an unsupervised learning method on student behavior
learning in the educational field [1], [2]. Several research based on learning log data generated from ubiquitous learning
studies in educational data mining and artificial intelligence applications. In this study, log data refers to all students’
have tempted to distinguish the LA movement in an educational activity while using the learning system namely ubiquitous
context [3], [4], [5], [6]. LA used educational data for fraction (U-Fraction) [14], [15]. This learning application is
knowledge discovery and transform data into meaningful installed on a tablet device with an Android operating system.
insights. It is used for leveraging educational data to support the By analyzing learning log data produced from the application,
teaching and learning process. The main purpose of LA is to the educational stakeholder can obtain learning problems at the
utilize models in improving learning and evaluating the process earliest possible time. Additionally, it can enable them to
 2
Journal of Advanced Technology and Multidiscipline (JATM)
Vol. 01, No. 01, Year, pp. 1-1x
resolve learning issues in a timelier fashion. Most importantly, their interaction behaviors rather than their perceptions and
a lot of data from learning systems and applications can be performance. The present study takes a further step toward the
analyzed using machine-learning techniques to support direction to propose an approach in interpreting students
decision-making in the educational field. We structured this learning log data to understand how students learn in the
paper as follows. In Section 1, an overview of LA especially for authentic situation over time. These findings are hinting that log
cluster analysis in an educational context was presented. data could be an important source to identify behavioral
Section 2 presented a literature review of related studies. interaction in authentic learning context.
Section 3 described the methodological part of the research.
B. Cluster analysis used in educational purpose
Furthermore, section 4 explained the result of the study
followed by a research discussion. Last, section 5 provided the To support the call for LA in education, several cluster
conclusion of the study. . analyses have been researched in the literature. While some
research studies have collected log-file data from virtual
II. LITERATURE REVIEW learning environments, the data was frequently evaluated using
more conventional statistical techniques like regression,
A. Learning log data in educational context correlation, and t-tests rather than analytics algorithms [22].
Learning log data is defined as important source to provide Instead, cluster analysis serves as an exploratory method that
powerful portrait of students learning patterns and their hidden aims to identify naturally occurring homogeneous groups that
behaviors during participation on learning process. Log data are were either unclear or previously unknown [23]. With a rapid
commonly collected from online learning platforms such as increase in available learner data, cluster analysis becomes the
virtual learning environment, e-learning, or mobile learning potential in understanding and unveiling hidden information
applications. Accessing and analyzing learning log data is about students in educational settings [24]. Studies by Yadav
challenging due to privacy issue and proper storage [25], [26] proposed a new approach known as hybrid clustering
management. Effective learning log data management requires to assess students’ academic performance. The clusters are
more time to be processed because the huge amount of formed based on the intelligence level of students. Walsh and
information collected from online server need complex Risquez [22] used cluster analysis to explore the engagement of
treatment like understanding of application usage, native and non-native English-speaking management students
preprocessing task, data engineering, and data architecture in a flipped classroom. They used log file data to identify hidden
provision. In the past research, some studies focused on the patterns in student behavior, paying particular attention to the
direction how to interpret learning log data in understanding institution's native language proficiency.
student learning process in flipped classroom [16], [17], [18]. Research shows the exploratory potential of cluster
Commonly, researchers on learning analytics used learning log analysis on log file data in other contexts such as peer tutoring
data from Learning Management System (e.g., Moodle, [27], [28]. However, despite its potential, cluster analysis is still
Canvas, etc.) or Massive Open Online Course (e.g., Coursera, underutilized in the context of education. Moreover, the rare
Udemy, etc.). The learning analytics goals emphasized teaching previous application of cluster analysis to study student
and learning processes in asynchronous learning networks. For learning behavior in ubiquitous learning contexts remained
example, data collection related to the number of posts, the unclear. The present study is adapted from the work of
number of posts read, the number of posts replied, and content Jovanovic et al. [29]. However, in the present study, we applied
viewed. cluster analysis to log-file data to identify patterns in how
A limitation of previous studies is that they focus on students access online resources over time while engaging with
student performance and student satisfaction which typically a ubiquitous learning application. This paper attempts to
rely on self-reporting and may be inaccurate [19], [20]. address the lack of research using learning analytics in the
Therefore, more studies based on log-file data are needed in ubiquitous learning context, using students' learning log data
order to add an additional level of research validity to the from a mobile application, and the cluster analysis algorithm
understanding of students' behavior in relation to the authentic using hierarchical and partitional methods.
learning approach. It has been argued that log-file data may be
more genuine and authentic than survey data, which are prone III. METHOD
bias into students' interpretations [21]. Instead, learning In the present study, we employed EDA as an initial
analytics can reflect real and uninterrupted user behavior [17]. technique for understanding the dataset. Investigation of data
Therefore, rather than relying on student perceptions, this study using EDA is used to discover unseen patterns, data anomalies,
examines ubiquitous application learning log data on how and a summary of the data [30]. Two important practices in
student interact in authentic learning context and how students EDA i.e., descriptive statistics and data visualization were used
accessed learning material. However, the use of learning log to gather insight from the data [31]. Before conducting EDA,
data in mobile application particularly for authentic learning we accessed the data from the online repository and organized
context had not yet fully exploited to unveil students learning it using Structured Query Language (SQL) operations such as
behaviors. Whereas, the adoption and acceptance of mobile data selection, data join, and data aggregation. We used
learning has led to a dramatic increase in available learner data. learning log data generated from a ubiquitous learning
Furthermore, students learning and social interaction with real- application namely U-Fraction. The dataset is related to student
world situation is critical to be learned and analyzed. Currently, learning activity while using the application such as problem-
there have been few studies that examine students' actions like solving activities and peer assessment. It was adapted from an
 3
Journal of Advanced Technology and Multidiscipline (JATM)
Vol. 01, No. 01, Year, pp. 1-1x
experimental study conducted by Hwang et. al. in 2018 [32]. each observation starts in its cluster, and pairs of clusters are
The data log structure before data preprocessing is represented merged as one moves up the hierarchy. This method works from
by the database design (Figure 2). There are 10 variables the dissimilarities between the objects to be grouped. A type of
selected for cluster analysis after the data preprocessing stage dissimilarity can be suited to the subject studied and the nature
and feature selection stage. The attributes of the dataset are of the data. Overall, the process in research methodology is
presented in Table 1. presented in Figure 1.
TABLE 1 TABLE 3
THE ATTRIBUTE OF THE DATASET K-MEANS PSEUDOCODE
No Attribute name Description Algorithm 1. K-Means Algorithm
1 Operation Total attempts of fraction operation Data: number of clusters k, dataset X
2 Success_oper Total of successful fraction operation Result: cluster centres C = {c1, ..., ck}
3 Simplification Total attempts at fraction simplification Start
4 Success_simp Total of successful fraction’s simplification Randomly select k data points as initial cluster centres;
5 Asking Problem posing Repeat
6 Answer Problem-solving Reinitialize all partition S subsets as empty:
7 Comment Peer assessment S1 = S2 = ··· = Sk = {};
8 Understanding Fraction understanding Compute the distance of each data point to each cluster centre;
9 Log1 Total data logging 1 Assign each data point to the closest cluster centre:
10 Log2 Total data logging 2 for i ∈ {1, ..., N} do
respective label l = argminj ∈ {1, ..., k} ‖ xi – cj ‖2;
Sl = Sl ∪ {xi};
After the data preprocessing step with EDA, we followed a two-
End
step cluster analysis using a K-means algorithm and Define new cluster centres based on the current partition:
agglomerative method. The K-means algorithm is a partition- for j ∈ {1, ..., k} do
based clustering method, while agglomerative is a hierarchical cj = ∑ i ∈ {1, ..., N} xi ∈ Sj xi / |Sj|
clustering method [27], [28]. K-means is best suited for a small- End
until the cluster assignment converges;
to-medium number of clusters, as is the case for student End
behavior clustering of this study [29]. The clustering process in
K-means started by defining the number of clusters k [30], [31]. IV. RESULT AND DISCUSSION
In addition, each of k is represented by a cluster center and each
data point is assigned to the nearest cluster center namely the In this section, we explained the results of the present
centroid. The algorithm group data that has similar study. The results of the study are categorized into two sub-
characteristics into the sample cluster, while data with different sections as follows:
characteristics are grouped into other clusters [32], [33]. A. Exploratory data analysis
Typically, the Euclidean distance is used as a distance measure.
In this step, we performed data pre-processing using EDA
The calculation using the Euclidian Distance formula (equation
such as data cleaning (i.e., missing value computation and data
1) with the description of the formula in Table 2 is as follows:
noise treatment), data transformation, and data reduction. EDA
d (x, y) = √∑𝑛𝑖=1(𝑦𝑖 − 𝑥𝑖 )2 (1) is important step in data analytic task because it performs initial
investigation on data to discover patterns, to spot some
TABLE 2 anomalies, to test hypothesis, and to check assumptions using
THE FORMULA DESCRIPTION
Symbol Description summary statistics and graphical representations. In present
d Calculation of the distance to the center of the cluster study, we employed several Python libraries such as Pandas,
x Point coordinates of the object NumPy, Matplotlib, and ScikitLearn to perform the EDA’s
y Centroid coordinate operation and cluster analysis. The learning log dataset
𝑛
The amount of data to be measured, while i = 1 is the comprises 4202 observations and 11 characteristics. We used
∑(𝑦𝑖 − 𝑥𝑖 )2
clustering process starting from the first iteration data.head(10) function to show the dataset with only ten rows
𝑖=1
xi Coordinate the point of the i object available (see Table 4). Furthermore, dataset information
yi i centroid coordinate point
including summary and missing value checking results is
presented in Figure 3. From the dataset summary, we can
In the next step, new cluster centers are defined as the center of identify the total of the column and the data type of each
mass of each cluster candidate. Unless the following column. Data has only non-null and integer values. In addition,
termination criterion is met, this process is repeated. The missing value analysis is used to check whether the dataset
algorithm terminates if the last iteration did not lead to changes contains a null value after the data pre-processing step. From
in the assignment of each data point to the current cluster the result, we concluded that all columns have no missing
centers [25]. The pseudocode is given in Table 3. Beside a K- values.
means algorithm, we also performed the agglomerative method
as a bottom-up approach to hierarchical clustering. Recursively,
 1
Journal of Advanced Technology and Multidiscipline (JATM)
Vol. 01, No. 01, Year, pp. 1-1x

Fig. 1 Research design

Fig. 2 The data log structures

Fig. 3 Dataset information. (a) data summary, (b) missing-value check

TABLE 4
THE DATASET WITH THE TOP 10 ROWS
User Operation Success_oper Simplify Success_simp Asking Answer Comment Understanding Log1 Log2
1 24 22 162 16 243 69 20 292 2621 518
2 45 44 128 20 242 74 14 287 2928 512
3 148 18 177 18 74 70 21 290 2601 541
4 9 38 138 44 97 73 12 258 2300 560
5 62 29 137 35 98 81 12 264 2382 1429
6 109 33 381 12 203 87 19 241 2290 3401
7 33 22 181 15 192 85 12 255 2892 506
8 26 18 239 12 107 93 12 291 2334 943
9 39 49 144 53 161 89 12 254 2284 567
10 48 20 155 16 153 135 12 244 2212 1418
 1
Journal of Advanced Technology and Multidiscipline (JATM)
Vol. 01, No. 01, Year, pp. 1-1x
TABLE 5
B. Cluster analysis with K-means algorithm and THE CLUSTER CHARACTERISTIC
agglomerative method Characteristic
Cluster 1 Cluster Cluster 3
(N=16) 2 (N=8) (N=1)
Cluster analysis in step 1, the student’s clusters of learning Total attempts of fraction operation 109.0 42.4 36.9
behavior are identified using a K-means algorithm. We used Total of successful fraction operation 33.0 22.9 26.0
two methods to select the optimum number of clusters k. The Total attempts at fraction 381.0 173.1 186.6
first method is based on Elbow Method, an empirical method to simplification
Total of successful fraction’s 12 22.7 22.8
obtain the best value of k. This method calculates the sum of simplification
the square of the points and the average distance. Figure 4 Problem-posing 203.0 158.0 125.9
shows the result of the elbow method. We concluded that the Problem-solving 87.0 99.1 96.6
optimal value of the cluster is 3 as presented by the last elbow Peer assessment 19.0 19.5 15.4
Fraction understanding 241.0 263.5 264.0
point (Figure 4). The second method is called the Silhouette Total data logging 1 2290.0 2426.4 2445.2
method. It calculates the silhouette coefficient of every point. Total data logging 2 3401.0 569.9 1501.1
The value of the Silhouette score varies from -1 to 1. Silhouette
score 1 means the cluster is dense and well-separated than other Based on the visualization, three clusters of students were
clusters. A value near 0 represents overlapping clusters with identified. Cluster 1 comprised more students with active
samples very close to the decision boundary of the neighboring learning behavior based on the total logs, total problems posed,
clusters. A negative score indicates that the samples might have and the total attempts in fraction operation and simplification.
got assigned to the wrong clusters. In this cluster analysis, we Students in clusters 2 and 3 had a higher attempt at problem-
obtained 3 clusters as the optimum value because it has a higher solving instead of problem-posing. Both clusters also focused
score of the Silhouette method (Figure 5). Furthermore, the on fraction understanding. Additionally, students in cluster 1
characteristic of each cluster is shown in Table 5. The result were similar to those in cluster 2 in terms of peer assessment
revealed different clusters of students based on their learning activity. In step 2 of cluster analysis, we performed hierarchical
behavior variable related to a ubiquitous learning activity. We clustering based on the agglomerative method. Figure 7 shows
also presented the comparison of each cluster of students’ a dendrogram, a diagram of the hierarchical relationship
learning behavior using parallel coordinates plots (Figure 6). between the students. The clades that are close to the same
height are similar to each other. The result revealed all students
of each cluster that has similar learning behavior characteristics.
According to the result, there are three clusters performed.
Cluster 1 consists of 16 students, followed by cluster 2 with 8
students and cluster 3 with 1 student. This result of hierarchical
clustering using the agglomerative method is similar to the
previous method using a K-means algorithm as partitional
clustering.

V. CONCLUSIONS
This research used the unsupervised learning method of
machine learning to discover a similar pattern of students’
learning log data and perform cluster analysis in order to obtain
students’ behavior clustering. The dataset is collected from
students’ learning activity while using the ubiquitous fraction
Fig. 4. Elbow method app called U-Fraction. Data are processed in the initial step
using EDA for data cleaning and transformation. Partition-
based clustering methods using the K-means algorithm and
hierarchical clustering methods using an agglomerative
approach are used to create a cluster of students. The result
showed three different clusters of students with different
learning behavior characteristics. Cluster 1 comprised more
students with active learning behavior based on the total logs,
total problems posed, and the total attempts in fraction
operation and simplification. Students in clusters 2 and 3 had a
higher attempt at problem-solving instead of problem-posing.
Both clusters also focused on fraction understanding. However,
no significant difference in peer assessment activity among the
groups. The outcome of this study can help educational
stakeholders to provide preventive learning strategies tailored
to different clusters of students.
Fig. 5 Silhouette score
 1
Journal of Advanced Technology and Multidiscipline (JATM)
Vol. 01, No. 01, Year, pp. 1-1x

Fig. 6 The comparison of each cluster

Fig. 7 Clustering results using hierarchical clustering

[3] A. A. Mubarak, H. Cao, and S. A. M. Ahmed, “Predictive learning

analytics using deep learning model in MOOCs’ courses videos,” Educ Inf
ACKNOWLEDGMENT Technol (Dordr), vol. 26, no. 1, pp. 371–392, 2021, doi: 10.1007/s10639-
All authors read and approved the final manuscript. 020-10273-6.
[4] M. Shorfuzzaman, M. S. Hossain, A. Nazir, G. Muhammad, and A. Alamri,
“Harnessing the power of big data analytics in the cloud to support learning
REFERENCES analytics in mobile learning environment,” Comput Human Behav, vol. 92,
[1] B. Quadir, M. Chang, and J. C. Yang, “Categorizing learning analytics pp. 578–588, 2019, doi: https://doi.org/10.1016/j.chb.2018.07.002.
models according to their goals and identifying their relevant components: [5] R. Bodily, R. Nyland, and D. Wiley, “The RISE Framework: Using
A review of the learning analytics literature from 2011 to 2019,” Learning Analytics to Automatically Identify Open Educational Resources
Computers and Education: Artificial Intelligence, vol. 2, p. 100034, 2021, for Continuous Improvement The RISE Framework: Using Learning
doi: https://doi.org/10.1016/j.caeai.2021.100034. Analytics to Automatically Identify Open Educational Resources for
[2] D. J. Lemay, C. Baek, and T. Doleck, “Comparison of learning analytics Continuous Improvement Bodily, Nyland, and Wiley,” 2017.
and educational data mining: A topic modeling approach,” Computers and [6] K. M. L. Jones, “Learning analytics and higher education: a proposed
Education: Artificial Intelligence, vol. 2, p. 100016, 2021, doi: model for establishing informed consent mechanisms to promote student
https://doi.org/10.1016/j.caeai.2021.100016. privacy and autonomy,” International Journal of Educational Technology
 2
Journal of Advanced Technology and Multidiscipline (JATM)
Vol. 01, No. 01, Year, pp. 1-1x
in Higher Education, vol. 16, no. 1, p. 24, 2019, doi: 10.1186/s41239-019- [24] W. Greller and H. Drachsler, “Translating Learning into Numbers: A
0155-0. Generic Framework for Learning Analytics,” 2012. [Online]. Available:
[7] J. Han, K. H. Kim, W. Rhee, and Y. H. Cho, “Learning analytics http://groups.google.com/group/learninganalytics
dashboards for adaptive support in face-to-face collaborative [25] J. Patel and R. S. Yadav, “Applications of Clustering Algorithms in
argumentation,” Comput Educ, vol. 163, p. 104041, 2021, doi: Academic Performance Evaluation,” OAlib, vol. 02, no. 08, pp. 1–14,
https://doi.org/10.1016/j.compedu.2020.104041. 2015, doi: 10.4236/oalib.1101623.
[8] M. T. Nafis and R. Biswas, “A secure technique for unstructured big data [26] R. S. Yadav, “Application of hybrid clustering methods for student
using clustering method,” International Journal of Information performance evaluation,” International Journal of Information Technology,
Technology, vol. 14, no. 3, pp. 1187–1198, 2022, doi: 10.1007/s41870- vol. 12, no. 3, pp. 749–756, 2020, doi: 10.1007/s41870-018-0192-2.
019-00278-x. [27] M. De Smet, H. Van Keer, and M. Valcke, “Blending asynchronous
[9] A. Shakarami, A. Shahidinejad, and M. Ghobaei-Arani, “An autonomous discussion groups and peer tutoring in higher education: An exploratory
computation offloading strategy in Mobile Edge Computing: A deep study of online peer tutoring behaviour,” Comput Educ, vol. 50, no. 1, pp.
learning-based hybrid approach,” Journal of Network and Computer 207–223, Jan. 2008, doi: 10.1016/j.compedu.2006.05.001.
Applications, vol. 178, p. 102974, 2021, doi: [28] G. Lust, M. Vandewaetere, E. Ceulemans, J. Elen, and G. Clarebout, “Tool-
https://doi.org/10.1016/j.jnca.2021.102974. use in a blended undergraduate course: In Search of user profiles,” Comput
[10] S. S. Hamidi, E. Akbari, and H. Motameni, “Consensus clustering Educ, vol. 57, no. 3, pp. 2135–2144, Nov. 2011, doi:
algorithm based on the automatic partitioning similarity graph,” Data 10.1016/j.compedu.2011.05.010.
Knowl Eng, vol. 124, p. 101754, 2019, doi: [29] J. Jovanović, D. Gašević, S. Dawson, A. Pardo, and N. Mirriahi, “Learning
https://doi.org/10.1016/j.datak.2019.101754. analytics to unveil learning strategies in a flipped classroom,” Internet and
[11] T. Li, A. Rezaeipanah, and E. M. Tag El Din, “An ensemble agglomerative Higher Education, vol. 33, pp. 74–85, Apr. 2017, doi:
hierarchical clustering algorithm based on clusters clustering technique and 10.1016/j.iheduc.2017.02.001.
the novel similarity measurement,” Journal of King Saud University - [30] A. K. Jain, R. P. W. Duin, and J. Mao, “Statistical pattern recognition: a
Computer and Information Sciences, vol. 34, no. 6, Part B, pp. 3828–3842, review,” IEEE Trans Pattern Anal Mach Intell, vol. 22, no. 1, pp. 4–37,
2022, doi: https://doi.org/10.1016/j.jksuci.2022.04.010. 2000, doi: 10.1109/34.824819.
[12] A. Mathrani, T. Susnjak, G. Ramaswami, and A. Barczak, “Perspectives [31] C.-M. Liu, Z. Niu, and K.-T. Liao, “Mechanisms to improve clustering
on the challenges of generalizability, transparency and ethics in predictive uncertain data with UKmeans,” Data Knowl Eng, vol. 116, pp. 61–79,
learning analytics,” Computers and Education Open, vol. 2, p. 100060, 2018, doi: https://doi.org/10.1016/j.datak.2018.05.004.
2021, doi: https://doi.org/10.1016/j.caeo.2021.100060. [32] W.-Y. Hwang, I. Q. Utami, S. W. D. Purba, and H. S. L. Chen, “Effect of
[13] S. Dumais, R. Jeffries, D. M. Russell, D. Tang, and J. Teevan, Ubiquitous Fraction App on Mathematics Learning Achievements and
“Understanding User Behavior Through Log Data and Analysis,” in Ways Learning Behaviors of Taiwanese Students in Authentic Contexts,” IEEE
of Knowing in HCI, J. S. Olson and W. A. Kellogg, Eds., New York, NY: Transactions on Learning Technologies, vol. 13, no. 3, pp. 530–539, 2020,
Springer New York, 2014, pp. 349–372. doi: 10.1007/978-1-4939-0378- doi: 10.1109/TLT.2019.2930045.
8_14. [33] N. Nidheesh, K. A. Abdul Nazeer, and P. M. Ameer, “An enhanced
[14] W.-Y. Hwang, I. Q. Utami, and H. Chen, “An Evaluation Study of deterministic K-Means clustering algorithm for cancer subtype prediction
Learning Behaviors and Achievements with Ubiquitous Fraction (u- from gene expression data,” Comput Biol Med, vol. 91, pp. 213–221, 2017,
Fraction) for Elementary School Student,” in 2018 IEEE 18th International doi: https://doi.org/10.1016/j.compbiomed.2017.10.014.
Conference on Advanced Learning Technologies (ICALT), 2018, pp. 350–
354. doi: 10.1109/ICALT.2018.00087. Ika Qutsiati Utami is a Ph.D. student at the Graduate
[15] I. Q. Utami and W.-Y. Hwang, “The impact of collaborative problem Institute of Network Learning Technology, College of
posing and solving with ubiquitous-decimal app in authentic contexts on Electrical Engineering & Computer Science, National
math learning,” Journal of Computers in Education, vol. 9, no. 3, pp. 427– Central University, ROC, Taiwan. She received B.S.
454, 2022, doi: 10.1007/s40692-021-00209-5. Degree in Information Systems from the Faculty of
[16] J. N. Walsh and A. Rísquez, “Using cluster analysis to explore the Computer Science, Brawijaya University, Indonesia &
engagement with a flipped classroom of native and non-native English- obtained M.S. Degree in Network Learning Technology
speaking management students,” International Journal of Management from the College of Electrical Engineering & Computer
Education, vol. 18, no. 2, Jul. 2020, doi: 10.1016/j.ijme.2020.100381. Science, National Central University, Taiwan. She is also working as a lecturer
[17] W. Greller and H. Drachsler, “Translating Learning into Numbers: A at the Faculty of Advanced Technology and Multidiscipline, Universitas
Generic Framework for Learning Analytics,” 2012. [Online]. Available: Airlangga, Indonesia. Her research interests are in the areas of Human-
http://groups.google.com/group/learninganalytics Computer Interaction, AI/ML for Future Education, and Educational Data
[18] D. Gašević, S. Dawson, T. Rogers, and D. Gasevic, “Learning analytics Science.
should not promote one size fits all: The effects of instructional conditions
in predicting academic success,” Internet and Higher Education, vol. 28, Wu-Yuin Hwang is a professor affiliated with both the
pp. 68–84, Jan. 2016, doi: 10.1016/j.iheduc.2015.10.002. Department of Computer Science and Information
[19] M. B. Gilboy, S. Heinerichs, and G. Pazzaglia, “Enhancing student Engineering, College of Science and Engineering,
engagement using the flipped classroom,” J Nutr Educ Behav, vol. 47, no. National Dong Hwa University, Taiwan, and the
1, pp. 109–114, Jan. 2015, doi: 10.1016/j.jneb.2014.08.008. Institute of Network Learning Technology, National
[20] J. Jovanović, D. Gašević, S. Dawson, A. Pardo, and N. Mirriahi, “Learning Central University, Taiwan. His current research
analytics to unveil learning strategies in a flipped classroom,” Internet and interests are related to the integration of IOT, AI, and
Higher Education, vol. 33, pp. 74–85, Apr. 2017, doi: multimedia sensors of mobile devices for learning and
10.1016/j.iheduc.2017.02.001. interactions among humans and all things in AR contexts like smart buildings
[21] I.-H. Jo, D. Kim, and M. Yoon, “International Forum of Educational and campuses. Dr. Hwang received the Outstanding Research Award, from the
Technology & Society Constructing Proxy Variables to Measure Adult Ministry of Science and Technology, Taiwan in 2021. He is also ranked in the
Learners’ Time Management Strategies in LMS,” Source: Journal of top 7 scholars of the world in terms of high-quality journal publication
Educational Technology & Society, vol. 18, no. 3, pp. 214–225, 2015, doi: performance of instructional design and technology.
10.2307/jeductechsoci.18.3.214.
[22] J. N. Walsh and A. Rísquez, “Using cluster analysis to explore the Ratih Ardiati Ningrum is a lecturer and researcher in
engagement with a flipped classroom of native and non-native English- Engineering Faculty at Airlangga University where she
speaking management students,” International Journal of Management has been a faculty member since 2020. She graduated
Education, vol. 18, no. 2, Jul. 2020, doi: 10.1016/j.ijme.2020.100381. with a Statistics master’s joint degree program from
[23] S. Brown, S. White, and N. Power, “Cluster Analysis of Assessment in National Chiao Tung University, Taiwan, and Institut
Anatomy and Physiology for Health Science Undergraduates,” Teknologi Sepuluh Nopember, Surabaya. Currently, her
International Journal of Teaching and Learning in Higher Education, vol. research focuses on survival analysis and structural
28, no. 1, pp. 102–109, 2016, [Online]. Available: equation modeling.
http://www.isetl.org/ijtlhe/