There is a growing need for student-centered learning (SCL) in higher education institutions, including dentistry. However, SCL has limited application in dental education. Therefore, this study aims to promote the application of SCL in dentistry by using decision tree machine learning (ML) technology to map the preferred learning style (LS) and corresponding learning strategies (IS) of dental students as a useful tool for developing IS guidelines. Promising methods for dental students.
A total of 255 dental students from the University of Malaya completed the modified Index of Learning Styles (m-ILS) questionnaire, which contained 44 items to classify them into their respective LSs. The collected data (called a dataset) is used in supervised decision tree learning to automatically match students’ learning styles to the most appropriate IS. The accuracy of the machine learning-based IS recommendation tool is then evaluated.
The application of decision tree models in an automated mapping process between LS (input) and IS (target output) allows for an immediate list of appropriate learning strategies for each dental student. The IS recommendation tool demonstrated perfect accuracy and recall of overall model accuracy, indicating that matching LS to IS has good sensitivity and specificity.
An IS recommendation tool based on an ML decision tree has proven its ability to accurately match dental students’ learning styles with appropriate learning strategies. This tool provides powerful options for planning learner-centered courses or modules that can enhance the learning experience of students.
Teaching and learning are fundamental activities in educational institutions. When developing a high-quality vocational education system, it is important to focus on the learning needs of students. The interaction between students and their learning environment can be determined through their LS. Research suggests that teacher-intended mismatches between students’ LS and IS can have negative consequences for student learning, such as decreased attention and motivation. This will indirectly affect student performance [1,2].
IS is a method used by teachers to impart knowledge and skills to students, including helping students learn [3]. Generally speaking, good teachers plan teaching strategies or IS that best match their students’ level of knowledge, the concepts they are learning, and their stage of learning. Theoretically, when LS and IS match, students will be able to organize and use a specific set of skills to learn effectively. Typically, a lesson plan includes several transitions between stages, such as from teaching to guided practice or from guided practice to independent practice. With this in mind, effective teachers often plan instruction with the goal of building students’ knowledge and skills [4].
The demand for SCL is growing in higher education institutions, including dentistry. SCL strategies are designed to meet students’ learning needs. This can be achieved, for example, if students actively participate in learning activities and teachers act as facilitators and are responsible for providing valuable feedback. It is said that providing learning materials and activities that are appropriate to students’ educational level or preferences can improve students’ learning environment and promote positive learning experiences [5].
Generally speaking, dental students’ learning process is influenced by the various clinical procedures they are required to perform and the clinical environment in which they develop effective interpersonal skills. The purpose of the training is to enable students to combine basic knowledge of dentistry with dental clinical skills and apply the acquired knowledge to new clinical situations [6, 7]. Early research into the relationship between LS and IS found that adjusting learning strategies mapped to the preferred LS would help improve the educational process [8]. The authors also recommend using a variety of teaching and assessment methods to adapt to students’ learning and needs.
Teachers benefit from applying LS knowledge to help them design, develop, and implement instruction that will enhance students’ acquisition of deeper knowledge and understanding of the subject matter. Researchers have developed several LS assessment tools, such as the Kolb Experiential Learning Model, the Felder-Silverman Learning Style Model (FSLSM), and the Fleming VAK/VARK Model [5, 9, 10]. According to the literature, these learning models are the most commonly used and most studied learning models. In the current research work, FSLSM is used to assess LS among dental students.
FSLSM is a widely used model for evaluating adaptive learning in engineering. There are many published works in the health sciences (including medicine, nursing, pharmacy and dentistry) that can be found using FSLSM models [5, 11, 12, 13]. The instrument used to measure the dimensions of LS in the FLSM is called the Index of Learning Styles (ILS) [8], which contains 44 items assessing four dimensions of LS: processing (active/reflective), perception (perceptual/intuitive), input (visual). /verbal) and understanding (sequential/global) [14].
As shown in Figure 1, each FSLSM dimension has a dominant preference. For example, in the processing dimension, students with “active” LS prefer to process information by directly interacting with learning materials, learn by doing, and tend to learn in groups. The “reflective” LS refers to learning through thinking and prefers to work alone. The “perceiving” dimension of LS can be divided into “feeling” and/or “intuition.” “Feeling” students prefer more concrete information and practical procedures, are fact-oriented compared to “intuitive” students who prefer abstract material and are more innovative and creative in nature. The “input” dimension of LS consists of “visual” and “verbal” learners. People with “visual” LS prefer to learn through visual demonstrations (such as diagrams, videos, or live demonstrations), whereas people with “verbal” LS prefer to learn through words in written or oral explanations. To “understand” the LS dimensions, such learners can be divided into “sequential” and “global”. “Sequential learners prefer a linear thought process and learn step by step, while global learners tend to have a holistic thought process and always have a better understanding of what they are learning.
Recently, many researchers have begun to explore methods for automatic data-driven discovery, including the development of new algorithms and models capable of interpreting large amounts of data [15, 16]. Based on the provided data, supervised ML (machine learning) is able to generate patterns and hypotheses that predict future results based on the construction of algorithms [17]. Simply put, supervised machine learning techniques manipulate input data and train algorithms. It then generates a range that classifies or predicts the outcome based on similar situations for the provided input data. The main advantage of supervised machine learning algorithms is its ability to establish ideal and desired results [17].
Through the use of data-driven methods and decision tree control models, automatic detection of LS is possible. Decision trees have been reported to be widely used in training programs in various fields, including health sciences [18, 19]. In this study, the model was specifically trained by the system developers to identify students’ LS and recommend the best IS for them.
The purpose of this study is to develop IS delivery strategies based on students’ LS and apply the SCL approach by developing an IS recommendation tool mapped to LS. The design flow of the IS recommendation tool as a strategy of the SCL method is shown in Figure 1. The IS recommendation tool is divided into two parts, including the LS classification mechanism using ILS and the most suitable IS display for students.
In particular, the characteristics of information security recommendation tools include the use of web technologies and the use of decision tree machine learning. System developers improve the user experience and mobility by adapting them to mobile devices such as mobile phones and tablets.
The experiment was carried out in two stages and students from the Faculty of Dentistry at the University of Malaya participated on a voluntary basis. Participants responded to a dental student’s online m-ILS in English. In the initial phase, a dataset of 50 students was used to train the decision tree machine learning algorithm. In the second phase of the development process, a dataset of 255 students was used to improve the accuracy of the developed instrument.
All participants receive an online briefing at the beginning of each stage, depending on the academic year, via Microsoft Teams. The purpose of the study was explained and informed consent was obtained. All participants were provided with a link to access the m-ILS. Each student was instructed to answer all 44 items on the questionnaire. They were given one week to complete the modified ILS at a time and location convenient to them during the semester break before the start of the semester. The m-ILS is based on the original ILS instrument and modified for dental students. Similar to the original ILS, it contains 44 evenly distributed items (a, b), with 11 items each, which are used to assess aspects of each FSLSM dimension.
During the initial stages of tool development, the researchers manually annotated the maps using a dataset of 50 dental students. According to the FSLM, the system provides the sum of answers “a” and “b”. For each dimension, if the student selects “a” as an answer, the LS is classified as Active/Perceptual/Visual/Sequential, and if the student selects “b” as an answer, the student is classified as Reflective/Intuitive/Linguistic. / global learner.
After calibrating the workflow between dental education researchers and system developers, questions were selected based on the FLSSM domain and fed into the ML model to predict each student’s LS. “Garbage in, garbage out” is a popular saying in the field of machine learning, with an emphasis on data quality. The quality of the input data determines the precision and accuracy of the machine learning model. During the feature engineering phase, a new feature set is created which is the sum of answers “a” and “b” based on FLSSM. Identification numbers of drug positions are given in Table 1.
Calculate the score based on the answers and determine the student’s LS. For each student, the score range is from 1 to 11. Scores from 1 to 3 indicate a balance of learning preferences within the same dimension, and scores from 5 to 7 indicate a moderate preference, indicating that students tend to prefer one environment teaching others. Another variation on the same dimension is that scores from 9 to 11 reflect a strong preference for one end or the other [8].
For each dimension, drugs were grouped into “active”, “reflective” and “balanced”. For example, when a student answers “a” more often than “b” on a designated item and his/her score exceeds the threshold of 5 for a particular item representing the Processing LS dimension, he/she belongs to the “active” LS domain. . However, students were classified as “reflective” LS when they chose “b” more than “a” in specific 11 questions (Table 1) and scored more than 5 points. Finally, the student is in a state of “equilibrium.” If the score does not exceed 5 points, then this is a “process” LS. The classification process was repeated for the other LS dimensions, namely perception (active/reflective), input (visual/verbal), and comprehension (sequential/global).
Decision tree models can use different subsets of features and decision rules at different stages of the classification process. It is considered a popular classification and prediction tool. It can be represented using a tree structure such as a flowchart [20], in which there are internal nodes representing tests by attribute, each branch representing test results, and each leaf node (leaf node) containing a class label.
A simple rule-based program was created to automatically score and annotate each student’s LS based on their responses. Rule-based takes the form of an IF statement, where “IF” describes the trigger and “THEN” specifies the action to be performed, for example: “If X happens, then do Y” (Liu et al., 2014). If the data set exhibits correlation and the decision tree model is properly trained and evaluated, this approach can be an effective way to automate the process of matching LS and IS.
In the second phase of development, the dataset was increased to 255 to improve the accuracy of the recommendation tool. The data set is split in a 1:4 ratio. 25% (64) of the data set was used for the test set, and the remaining 75% (191) was used as the training set (Figure 2). The data set needs to be split to prevent the model from being trained and tested on the same data set, which could cause the model to remember rather than learn. The model is trained on the training set and evaluates its performance on the test set—data the model has never seen before.
Once the IS tool is developed, the application will be able to classify LS based on the responses of dental students via a web interface. The web-based information security recommendation tool system is built using the Python programming language using the Django framework as the backend. Table 2 lists the libraries used in the development of this system.
The dataset is fed to a decision tree model to calculate and extract student responses to automatically classify student LS measurements.
The confusion matrix is used to evaluate the accuracy of a decision tree machine learning algorithm on a given data set. At the same time, it evaluates the performance of the classification model. It summarizes the model’s predictions and compares them to the actual data labels. The evaluation results are based on four different values: True Positive (TP) – the model correctly predicted the positive category, False Positive (FP) – the model predicted the positive category, but the true label was negative, True Negative (TN) – the model correctly predicted the negative class, and false negative (FN) – The model predicts a negative class, but the true label is positive.
These values are then used to calculate various performance metrics of the scikit-learn classification model in Python, namely precision, precision, recall, and F1 score. Here are examples:
Recall (or sensitivity) measures the model’s ability to accurately classify a student’s LS after answering the m-ILS questionnaire.
Specificity is called a true negative rate. As you can see from the above formula, this should be the ratio of true negatives (TN) to true negatives and false positives (FP). As part of the recommended tool for classifying student drugs, it should be capable of accurate identification.
The original dataset of 50 students used to train the decision tree ML model showed relatively low accuracy due to human error in the annotations (Table 3). After creating a simple rule-based program to automatically calculate LS scores and student annotations, an increasing number of datasets (255) were used to train and test the recommender system.
In the multiclass confusion matrix, the diagonal elements represent the number of correct predictions for each LS type (Figure 4). Using the decision tree model, a total of 64 samples were correctly predicted. Thus, in this study, the diagonal elements show the expected results, indicating that the model performs well and accurately predicts the class label for each LS classification. Thus, the overall accuracy of the recommendation tool is 100%.
The values of accuracy, precision, recall, and F1 score are shown in Figure 5. For the recommendation system using the decision tree model, its F1 score is 1.0 “perfect,” indicating perfect precision and recall, reflecting significant sensitivity and specificity values.
Figure 6 shows a visualization of the decision tree model after training and testing are completed. In a side-by-side comparison, the decision tree model trained with fewer features showed higher accuracy and easier model visualization. This shows that feature engineering leading to feature reduction is an important step in improving model performance.
By applying decision tree supervised learning, the mapping between LS (input) and IS (target output) is automatically generated and contains detailed information for each LS.
The results showed that 34.9% of the 255 students preferred one (1) LS option. The majority (54.3%) had two or more LS preferences. 12.2% of students noted that LS is quite balanced (Table 4). In addition to the eight main LS, there are 34 combinations of LS classifications for University of Malaya dental students. Among them, perception, vision, and the combination of perception and vision are the main LS reported by students (Figure 7).
As can be seen from Table 4, the majority of students had a predominant sensory (13.7%) or visual (8.6%) LS. It was reported that 12.2% of students combined perception with vision (perceptual-visual LS). These findings suggest that students prefer to learn and remember through established methods, follow specific and detailed procedures, and are attentive in nature. At the same time, they enjoy learning by looking (using diagrams, etc.) and tend to discuss and apply information in groups or on their own.
This study provides an overview of machine learning techniques used in data mining, with a focus on instantly and accurately predicting students’ LS and recommending suitable IS. Application of a decision tree model identified the factors most closely related to their life and educational experiences. It is a supervised machine learning algorithm that uses a tree structure to classify data by dividing a set of data into subcategories based on certain criteria. It works by recursively dividing the input data into subsets based on the value of one of the input features of each internal node until a decision is made at the leaf node.
The internal nodes of the decision tree represent the solution based on the input characteristics of the m-ILS problem, and the leaf nodes represent the final LS classification prediction. Throughout the study, it is easy to understand the hierarchy of decision trees that explain and visualize the decision process by looking at the relationship between input features and output predictions.
In the fields of computer science and engineering, machine learning algorithms are widely used to predict student performance based on their entrance exam scores [21], demographic information, and learning behavior [22]. Research showed that the algorithm accurately predicted student performance and helped them identify students at risk for academic difficulties.
The application of ML algorithms in the development of virtual patient simulators for dental training is reported. The simulator is capable of accurately reproducing the physiological responses of real patients and can be used to train dental students in a safe and controlled environment [23]. Several other studies show that machine learning algorithms can potentially improve the quality and efficiency of dental and medical education and patient care. Machine learning algorithms have been used to assist in the diagnosis of dental diseases based on data sets such as symptoms and patient characteristics [24, 25]. While other studies have explored the use of machine learning algorithms to perform tasks such as predicting patient outcomes, identifying high-risk patients, developing personalized treatment plans [26], periodontal treatment [27], and caries treatment [25].
Although reports on the application of machine learning in dentistry have been published, its application in dental education remains limited. Therefore, this study aimed to use a decision tree model to identify factors most closely associated with LS and IS among dental students.
The results of this study show that the developed recommendation tool has high accuracy and perfect accuracy, indicating that teachers can benefit from this tool. Using a data-driven classification process, it can provide personalized recommendations and improve educational experiences and outcomes for educators and students. Among them, information obtained through recommendation tools can resolve conflicts between teachers’ preferred teaching methods and students’ learning needs. For example, due to the automated output of recommendation tools, the time required to identify a student’s IP and match it with the corresponding IP will be significantly reduced. In this way, suitable training activities and training materials can be organized. This helps develop students’ positive learning behavior and ability to concentrate. One study reported that providing students with learning materials and learning activities that match their preferred LS can help students integrate, process, and enjoy learning in multiple ways to achieve greater potential [12]. Research also shows that in addition to improving student participation in the classroom, understanding students’ learning process also plays a critical role in improving teaching practices and communication with students [28, 29].
However, as with any modern technology, there are problems and limitations. These include issues related to data privacy, bias and fairness, and the professional skills and resources needed to develop and implement machine learning algorithms in dental education; However, growing interest and research in this area suggests that machine learning technologies may have a positive impact on dental education and dental services.
The results of this study indicate that half of dental students have a tendency to “perceive” drugs. This type of learner has a preference for facts and concrete examples, a practical orientation, patience for detail, and a “visual” LS preference, where learners prefer to use pictures, graphics, colors, and maps to convey ideas and thoughts. The current results are consistent with other studies using ILS to assess LS in dental and medical students, most of whom have characteristics of perceptual and visual LS [12, 30]. Dalmolin et al suggest that informing students about their LS allows them to reach their learning potential. Researchers argue that when teachers fully understand students’ educational process, various teaching methods and activities can be implemented that will improve students’ performance and learning experience [12, 31, 32]. Other studies have shown that adjusting students’ LS also shows improvements in students’ learning experience and performance after changing their learning styles to suit their own LS [13, 33].
Teachers’ opinions may vary regarding the implementation of teaching strategies based on students’ learning abilities. While some see the benefits of this approach, including professional development opportunities, mentorship, and community support, others may be concerned about time and institutional support. Striving for balance is key to creating a student-centered attitude. Higher education authorities, such as university administrators, can play an important role in driving positive change by introducing innovative practices and supporting faculty development [34]. To create a truly dynamic and responsive higher education system, policymakers must take bold steps, such as making policy changes, devoting resources to technology integration, and creating frameworks that promote student-centered approaches. These measures are critical to achieving the desired results. Recent research on differentiated instruction has clearly shown that successful implementation of differentiated instruction requires ongoing training and development opportunities for teachers [35].
This tool provides valuable support to dental educators who want to take a student-centered approach to planning student-friendly learning activities. However, this study is limited to the use of decision tree ML models. In the future, more data should be collected to compare the performance of different machine learning models to compare the accuracy, reliability, and precision of recommendation tools. Additionally, when choosing the most appropriate machine learning method for a particular task, it is important to consider other factors such as model complexity and interpretation.
A limitation of this study is that it only focused on mapping LS and IS among dental students. Therefore, the developed recommendation system will only recommend those that are suitable for dental students. Changes are necessary for general higher education student use.
The newly developed machine learning-based recommendation tool is capable of instantly classifying and matching students’ LS to the corresponding IS, making it the first dental education program to help dental educators plan relevant teaching and learning activities. Using a data-driven triage process, it can provide personalized recommendations, save time, improve teaching strategies, support targeted interventions, and promote ongoing professional development. Its application will promote student-centered approaches to dental education.
Gilak Jani Associated Press. Match or mismatch between the student’s learning style and the teacher’s teaching style. Int J Mod Educ Computer Science. 2012;4(11):51–60. https://doi.org/10.5815/ijmecs.2012.11.05
Post time: Apr-29-2024