Competent medical students are essential for cultivating a skilled cadre of doctors who not only contribute directly to patient healthcare but also enhance collaboration within the national public health system.

To achieve these goals, most medical institutions strive to develop optimal educational programs that achieve academic success and professionalism. Addressing the complexity of multifactorial data is a critical challenge in this process. As a result, the introduction of educational data mining (EDM) has gained attention as a promising solution.

EDM is an emerging interdisciplinary research area that involves education and computer science, with the ability to extract and convert large amounts of educational data into useful patterns of information and important knowledge. Various tasks are involved, such as description, prediction, estimation, classification, clustering, and association discovering, leading to the development of numerous techniques, including regression, naïve Bayes, decision trees, neural networks (NN), K-means, Apriori, FP-Growth, and many more (Aleem and Gore, 2020). Until now, several applications of EDM have emerged, such as regression models for graduate-level performance prediction (Zimmermann et al., 2015), association rule mining for teaching assessment (Sy and Que, 2019), and cluster analysis for learning style classification (Beal et al., 2006).

While various definitions of student success exist in the literature, they focus primarily on academic achievements and professionalism, and the latter is difficult to measure (Kuh et al., 2006; Alyahyan and Düştegör, 2020). Student achievement is commonly addressed using the grade point average (GPA), which remains a crucial consideration for postgraduate training programs such as residency and fellowship (McGaghie, 2019).

Numerous studies in the literature have explored influential factors based on learning, cognitive, and behavioral theories for predicting academic success (Kolb, 2014). Our research primarily concentrates on four domains: academic activity, demographics, environment, and psychology or learning style. These domains are frequently reported as predictive or associated with our outcomes in numerous systematic literature reviews (Alyahyan and Düştegör, 2020; Baashar and Alkawsi, 2021; Hellas et al., 2018; Crisp et al., 2015), with prior academic achievement and demographics ranking highest (44 and 25%, respectively), followed by environment and psychology or learning style (17 and 11%, respectively) (Alyahyan and Düştegör, 2020). Psychology, or learning style, is defined as the composite of cognitive, affective, and psychological characteristics that are stable indicators of how a learner perceives, interacts with, and responds to the learning environment (Nguyen, 2013). Unfortunately, a multifactorial linkage is believed to occur due to the complexity of human behavior and the learning process (Alyahyan and Düştegör, 2020; Salem et al., 2013).

Compared to traditional medical education, which relies on passive memorization of content, recent research has introduced more active approaches, emphasizing creativity and innovative active learning (Hindin et al., 2023).

Notably, educational analysis could engage with many dimensions and data types, such as categorical data in attitude or learning style and continuous GPA data. Thus, it is a unique tool that can potentially be applied to collect and mine different data types for targeted outcomes.

Definition: Student success is defined as academic achievement, engagement in educationally purposeful activities, satisfaction, acquisition of desired knowledge, skills and competencies, persistence, attainment of educational outcomes, and post-college performance (Kuh et al., 2006).

The retrospective cohort study was conducted using an online standardized questionnaire consisting of 13 questions and one learning style test (Table 1). The questionnaire was administered to medical students who graduated from Phramongkutklao College of Medicine between April 2010 and April 2023. It included questions pertaining to specific periods of the 6-year medical student experience and was conducted from January 2022 to April 2023. The study followed a four-step approach, modified from the Foundational Method for Data Science (FMD), IBM 2015, as presented in Figure 1.

The primary outcome was academic success, measuring the accumulated grade point average of graduated students in medical school, or GPAX (Med). The factors affecting our outcomes were analyzed from the literature (Alyahyan and Düştegör, 2020; Salem et al., 2013) and concluded in four domains: academic activity, demographics, environment, and learning style or psychology. Based on students’ data, the inclusion criteria were graduation between April 2010 and April 2023 to minimize the recall bias and ensure the voluntary completion of all questionnaire questions. Incomplete questionnaires were excluded. The Royal Thai Army Medical Department Institutional Review Board approved the study, reference number R192h/64_Exp. Student identification was coded before analysis and discussion.

Study population: a dataset of 145 medical students.

Attributes (study factors): a total of 13 attributes from four categorical and numerical data domains, as shown in Table 1. All factors were selected based on their statistical and clinical relevance to our research outcomes from literature reviews.

Data preparation:

Logistic regression: prevalence ratios and adjusted prevalence ratios were calculated using univariate and multivariate logistic regression, with statistical significance determined by a p-value <0.05.

Model training–testing (internal validation)/deployment (external validation): further research with a new cohort is required to test the model using an independent sample, as our study’s datasets are inadequate.

Our study successfully employed logistic regression, cluster analysis, and association rules to identify factors affecting academic success among medical students, with a particular focus on visualization techniques; it is illustrated in a tree framework (Figure 5). Each method provided unique insights, enhancing our understanding of how academic, demographic, environmental, and psychological factors interact to shape student outcomes.

Logistic regression was applied to identify significant predictors of academic performance, with a pre-med GPAX of ≥ 3.75 and an interest in internal medicine emerging as statistically significant factors associated with higher grades (APR 1.73, p = 0.005; APR 1.52, p = 0.040, respectively). Margins analysis also showed a predicted mean GPAX (med) of 3.41 (p < 0.005) for students with a pre-med GPAX of ≥3.75. This finding aligns with previous research highlighting GPAX as a strong predictor of academic performance (Furnham and Monsen, 2009; Jacobparayil et al., 2022). While baseline GPAX is unchangeable, ongoing support and targeted interventions could benefit students with lower pre-med GPAX, emphasizing the need for early awareness and academic assistance. Additionally, motivation toward specialty choice was found to influence academic outcomes, with internal medicine showing a significant association. Strong correlations were observed between internal medicine and an intellectual, inquisitive, and ‘investigative’ personality profile, as identified through a validated 140-item questionnaire in a cluster analysis study (Sievert et al., 2016). This essentially suggests that the personality traits that draw students to internal medicine—such as intellectual curiosity and a preference for investigation—might also help them succeed in the cognitive assessments commonly used in medical education. This alignment could help explain why students interested in internal medicine tend to perform better academically. Similar associations between competitive specialties and academic performance have also been noted (Mitsouras et al., 2019). This may also reflect a supply–demand dynamic, where students motivated to pursue competitive fields with high academic standards work harder to achieve better grades. However, we recognize that personal preferences, experiences, and external factors can influence specialty choice and academic outcomes (Soethout et al., 2008). The above findings positively show the important role of factors in “academic activities.” However, we found no statistical relevance to demographic, environmental, or psychological factors. Interestingly, the finding of irrelevance of learning style and high-grade students, referring to high-performing students, indicates that they may have different ways of learning. Similar results have been reported in studies involving dental and medical students, including a larger cohort of 600 medical students (Mozaffari et al., 2020; Almigbal, 2015). Notably, subgroup analysis found that 40% (14 in 35) of high-grade students who were interested in medicine had a K learning style and achieved high grades. Similarly, other studies conducted involving medical students from Thailand and Saudi Arabia have suggested that specific learning styles are associated with higher achievement. Those were sequential learning or logical sequencing patterns and visuality (Jiraporncharoen et al., 2015; Al Fhaid et al., 2022). These contrasting results highlight the need for teachers to recognize that “everyone cannot be taught the same way” when designing classroom interventions and environments (Shakeri et al., 2022).

However, we acknowledge that focusing solely on these two factors might oversimplify the complexity of academic success, as other potential influences were excluded due to a lack of statistical significance. This limitation indicates the need for additional methods to capture multidimensional factors beyond simple predictive models.

Utilizing cluster analysis to address the limitations of LR, we employed cluster analysis on unlabeled data, grouping students based on shared characteristics without predefined academic performance labels. This exploratory approach allowed us to uncover patterns within each cluster and then observe performance trends. We could identify typical characteristics of the high-grade group, which included residing in a metropolis, achieving a very high GPAX during pre-med, and a strong preference for kinesthetic learning, followed by the reading learning style. While the first two factors are non-modifiable, the discovery of kinesthetic learning style as a key factor in high-grade success is noteworthy. In contrast to the lower grade group, various types with slightly higher reading styles but very low K-style were investigated. This revealed a very high-grade cluster (GPAX centroid of 3.67), and K-style might be an interesting key factor in its success. As medical programs require both knowledge and skill, we assume that students approach their learning through a combination of reading and hands-on learning to effectively prepare students for the medical curriculum and lead to an honors degree. This aligns with the concept that medicine is not solely about scientific knowledge but also knowledge-driven practice or kinesthetic skill (De Camargo, and Jr. Coeli, 2006). However, this theory should be applied cautiously, as observational studies using questionnaire reports that study strategies based on VARK results showed no correlation with anatomy course outcomes (Husmann and O'Loughlin, 2019).

Finally, association analysis discovered informative rules emphasizing the importance of transportation time to school and access to learning resources, highlighting the significance of institutional resources in facilitating student knowledge acquisition.

Exploring individual learning styles can empower students to understand their strengths and preferences, aligning with Howard Gardner’s theory of multiple intelligences, which emphasizes that learners excel through different modalities (Durie, 1997; Hopper and Hurry, 2000). Recognizing one’s learning style can encourage self-directed learning, as outlined in constructivist theories, where students actively build understanding through engagement with content and adapt to the academic environment (Vygotsky, Piaget’s theories) (Blake, 2008). Incorporating learning styles into lesson planning may enhance students’ ability to relate to and comprehend material, fostering deeper and faster practical application. This aligns with Kolb’s experiential learning cycle, specifically the active experimentation phase, where learners apply new knowledge to real-world scenarios, cultivating skills for lifelong learning (Kolb, 2014).

For Teachers: We advocate for a tailored, student-centered approach to address diverse learning needs.

Similar to colors, students possess unique “hues” or learning styles, and by recognizing these differences, educators can implement varied instructional strategies to support and actualize each student’s potential. Howard Gardner’s theory of multiple intelligences supports this perspective, encouraging instructors to pluralize their teaching and assessment methods, thereby enriching student learning by addressing multiple modalities (Durie, 1997; Hopper and Hurry, 2000). Teachers can make learning more effective and inclusive by investing time to understand each learner’s style and implementing multi-modality instructional methods, from passive (reading, auditory lectures) to active (kinesthetic activities, simulations, and role-play).

Dale’s Cone of Experience reinforces this approach, demonstrating that hands-on and participatory methods often lead to better retention and understanding (Dale, 1969; Amin and Khoo, 2003). This strategy aligns with constructivist and humanistic theories, which emphasize active, collaborative, and context-rich learning for self-actualization. For example, problem-based learning (PBL) encourages students to engage in real-world problem-solving, deepening their understanding. Humanistic theories also advocate for a nurturing environment that fosters personal development and self-determination, underscoring the importance of adaptive, flexible teaching strategies (Blake, 2008; Brockett, 1996). Additionally, connectivism suggests that students in the digital age benefit from continually updating knowledge and connecting new and existing information, making a pluralistic approach even more relevant (Duke et al., 2013).

Finally, institutions are pivotal in supporting student success by modifying environmental factors. Reducing transportation time and enhancing access to learning resources, such as digital tools and quiet study spaces, can positively impact academic outcomes. Maslow’s hierarchy of needs suggests that meeting students’ basic needs (e.g., safe and accessible study environments) can facilitate higher-order achievements such as academic success. This aligns with humanistic theory, which highlights the role of a supportive and nurturing environment in fostering learning. When students have access to essential resources, they are more likely to engage fully with their studies, potentially achieving self-actualization as lifelong learners (Brockett, 1996).

The strength of our study lies in the introduction of two unique analytical techniques—logistic regression and data mining—which enabled us to identify both statistically significant factors and meaningful hidden patterns. To the best of our knowledge, this is the first research to use multidomain factors to investigate medical students’ academic success, analyzing them with these techniques and emphasizing visualization.

Our study has several limitations. As a retrospective study relying on medical students’ recollections of experiences, recall bias may be present. We sought to reduce this by limiting the number of recall questions, focusing on specific periods within the 6-year medical program to improve accuracy. Additionally, reliance on self-reported data introduces potential recall and social desirability biases. To mitigate recall bias, we designed clear and specific questions, especially for demographic and environmental factors, and pre-tested the questionnaire to enhance reliability. To reduce social desirability bias, we assured participants of confidentiality and included disclaimers to encourage honesty. Another limitation relates to our sample composition. The study’s focus and title may have attracted high-achieving students, potentially skewing the prevalence of high GPAX students in our sample and impacting representativeness. However, as our primary objective was to examine associations between factors, we believe this limitation only minimally affects the findings. Another limitation that could confound the outcome was the variation of students’ graduation years and other uncontrolled external factors such as scoring, teamwork, or health issues. However, multivariate analysis using a session period as a training course was believed to eliminate one confounder. Noteworthy that our populations might not truly represent the general medical student population, and further research is planned to investigate different universities and regions of the country. Future research may benefit from larger, more diverse datasets to enhance robustness and enable external validation, supporting broader generalizability. Additionally, advanced methods like hierarchical clustering and SVMs could uncover nuanced patterns and complex relationships, providing further insights for educational programs and decision-making.

By utilizing three analytical methods, we identified key factors influencing medical students’ academic success. Logistic regression highlighted pre-med GPAX and an interest in internal medicine as significant predictors. Cluster analysis revealed distinct learning style patterns, with high-grade students favoring kinesthetic and reading styles, while lower-grade students preferred auditory and visual styles. Association rules emphasized the role of environmental factors, such as proximity to school and access to resources, in supporting academic outcomes. Together, these findings underscore the importance of a holistic approach in educational planning that considers learning preferences, personal interests, and supportive environments.

In summary, a multi-domain framework in educational planning can foster more effective and personalized learning experiences. By addressing individual learning styles, adaptive teaching strategies, and supportive environments, institutions can better equip medical students to achieve their academic and professional goals.