IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 14, No. 5, October 2025, pp. 4113~4122

ISSN: 2252-8938, DOI: 10.11591/ijai.v14.i5.pp4113-4122  4113

Educational data mining approach for predicting student

performance and behavior using deep learning techniques

Muniappan Ramaraj1, Sabareeswaran Dhendapani2, Jothish Chembath3, Selvaraj Srividhya4,

Nainan Thangarasu5, Bhaarathi Ilango1
1
Department of Computer Science, Rathinam College of Arts and Science, Coimbatore, India
2
Department of Computer Science, L.R.G. Government College of Arts and Science for Women, Tiruppur, India
3
Department of Computer Science, Presidency University, Bengaluru, lndia
4
Department of Computer Science, KPR College of Arts Science and Research, Coimbatore, India
5
Department of Computer Science, Karpagam Academy of Higher Education, Coimbatore, India

Article Info ABSTRACT

Article history: Educational data mining (EDM) uncovers insights from large datasets
collected from various educational platforms, such as online learning
Received Sep 28, 2024 systems, student information databases, and classroom tools. EDM helps
Revised Jun 27, 2025 educators identify hidden patterns that improve teaching strategies,
Accepted Aug 6, 2025 personalize learning experiences, and predict student performance.
Predicting student success has become a key focus of EDM, allowing
institutions to implement targeted interventions and personalized support.
Keywords: The dataset included academic achievement grades from 1,001 students
enrolled in various courses during the fall semester across multiple years, to
Artificial neural networks demonstrate how proposed models provide more accurate predictions
Deep learning compared to traditional machine learning methods. Models such as you only
Educational data mining look once (YOLO), fast region-based convolutional neural networks (Fast
Long short-term memory RCNN), artificial neural networks (ANNs), and long short-term memory
Predictive analytics (LSTM) networks are used to capture complex, non-linear relationships
Student performance prediction within the data. The comparative analysis shows that these deep learning
models significantly outperform traditional techniques, such as decision
trees and support vector machines (SVMs). The results indicate that
proposed method offers improved predictive accuracy, enabling educational
institutions to identify at-risk students and deliver tailored interventions.
This study highlights the potential of enhanced method to transform
personalized education and enhance student success by better understanding
individual learning needs and behaviors.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Muniappan Ramaraj
Department of Computer Science, Rathinam College of Arts and Science
Coimbatore, India
Email: [email protected]

1. INTRODUCTION
Educational data mining (EDM) is a growing field dedicated to uncovering valuable patterns and
insights from extensive educational datasets. By effectively utilizing this data, institutions can forecast
student outcomes, detect students at risk, and customize educational interventions to meet individual needs
[1]. Recent progress in machine learning and deep learning has led to the creation of highly accurate models
for predicting student performance [2]. Although traditional machine learning models, such as decision trees
and support vector machines (SVMs), have been applied to this task, deep learning techniques have
demonstrated significant potential due to their capacity to handle large datasets and uncover intricate,

Journal homepage: http://ijai.iaescore.com

4114  ISSN: 2252-8938

non-linear relationships [3]. This paper seeks to investigate the application of deep learning models for
predicting student performance, with a particular emphasis on artificial neural networks (ANNs) and long
short-term memory (LSTM) networks [4]. In many virtual learning environments (VLEs), however, datasets
can be sparse or imbalanced, making simpler models such as decision trees, logistic regression, or even
ensemble methods like random forests more practical in real-world applications [5].
Learning analytics (LA) typically focuses on real-time, actionable insights for immediate
interventions, while EDM emphasizes discovering patterns in large datasets through data mining techniques.
The problem arises when attempting to combine the real-time focus of LA with the data-driven discovery of
EDM, as each operates on different temporal and methodological planes [6]. A deep cognitive diagnosis
model (DCDM) for predicting students’ performance focuses on enhancing how accurately we can assess
student knowledge based on their responses to various assessments [7]. Another issue is the need for large-
scale, high-quality data to train such models effectively. In many educational contexts, data may be scarce,
noisy, or imbalanced, especially in terms of assessments for specific learning domains or minority student
groups [8]. Many deep learning-based knowledge tracing (DLKT) models are trained on specific types of
data (e.g., online learning platforms and standardized tests). The survey seeks to address the problem of how
well DLKT models generalize across diverse learning contexts. However, the problem remains that there is
limited research exploring the direct role artificial intelligence (AI) can play in enhancing academic outcomes
by focusing on both study strategies and learning disabilities [9]. A novel machine learning model, random
grouping-based deep multi-modal learning (RG-DMML), which is coupled with an ensemble learning
algorithm. This model integrates various data sources, such as academic records and demographic
information, and applies deep learning techniques to enhance prediction accuracy [10]. Educational
institutions struggle to identify at-risk students early enough to intervene effectively.

2. LITERATURE REVIEW
EDM has seen rapid growth over the last decade, driven by the increasing availability of educational
data from online learning platforms, student information systems, and other digital tools. Early research
focused on rule-based systems and statistical models, which, while effective in certain scenarios, struggled to
scale with increasing data complexity. Rathi et al. [11] has presents the hybrid approach combining the self-
supervised robust optimization algorithm (SS-ROA) and deep LSTM networks. This model leverages the
strengths of deep learning in handling time-series data while optimizing feature selection and model training
using the SS-ROA technique. Ding [12] has illustrate on deep learning models can analyze student
movements through video data, providing real-time corrections or feedback on technique and posture. In
music, AI-driven models can assess pitch, timing, and expression during performances, offering students
detailed feedback on areas for improvement. Aulakh et al. [13] aims to examine the intersection of e-learning
and EDM during the COVID-19 pandemic. It explores various EDM methods applied in e-learning, such as
clustering, classification, and regression analysis.
Sarker et al. [14] analyzing students’ academic performance through EDM has emerged as a
valuable approach for improving educational outcomes and institutional decision-making. Feng and Fan [15]
has investigate how EDM can improve the learning process by evaluating learning behaviors, predicting
student success, and visualizing data in a way that supports decision-making in education. Deng et al. [16]
has introduces a novel deep learning-based predictive model, capable of analyzing various factors such as
self-esteem levels, tendencies towards individualism, and their combined impact on performance metrics.
Lam et al. [17] introduces a robust framework that leverages machine learning techniques to accurately
predict student performance, enabling proactive identification of learners at academic risk. By utilizing
algorithms such as k-means, hierarchical clustering, and density-based spatial clustering of applications with
noise (DBSCAN), the study seeks to uncover patterns that can inform educators about the diverse needs of
their students. Peng et al. [18] the achievement of this research lies in its ability to facilitate targeted
interventions, personalized learning pathways, and ultimately enhance educational outcomes.
Rejeb et al. [19] aims to examine how ChatGPT is being utilized in various educational contexts and
to assess its influence on teaching methods, learning experiences, and overall educational outcomes.
Bhardwaj et al. [20] demonstrates that deep learning models, such as convolutional neural networks (CNNs)
and LSTM networks, are effective tools for predicting and analyzing student engagement in e-learning
environments. It aims to identify patterns of engagement, predict student behaviors, and provide personalized
interventions to improve learning outcomes. Al Ka'bi [21] has introduces a novel AI algorithm and deep
learning techniques tailored for enhancing the quality of higher education. Lin et al. [22] aims to streamline
learning processes, improve educational outcomes, and optimize institutional management by providing
personalized learning experiences, predictive analytics, and automated administrative tasks. Farhood et al. [23]
contributes to the field of EDM by introducing generative adversarial networks (GANs) as a novel approach

Int J Artif Intell, Vol. 14, No. 5, October 2025: 4113-4122

Int J Artif Intell ISSN: 2252-8938  4115

for improving student outcome predictions. The focus is on extracting meaningful patterns and insights from
textual or communication data generated during learning processes, such as online discussions, written
assignments, or feedback. Riaz et al. [24] introduces TransLSTM, a novel hybrid architecture combining the
strengths of LSTM and Transformer models to perform fine-grained suggestion mining.

3. METHOD
The methodology for predicting student performance through EDM, this study employs a multi-step
methodology utilizing deep learning techniques [25]. The approach begins with data collection, where
academic records, demographic details, and behavioral patterns are aggregated. The data undergoes
preprocessing to clean and normalize it, followed by feature selection to identify the most relevant attributes
for prediction [26].
Figure 1 illustrates on the predicting student performance using deep learning in EDM involves
several key steps. Initially, a dataset comprising academic records, demographic details, and behavioral data
is collected and preprocessed to handle missing values and normalize features. To identify the most relevant
features for prediction, feature selection is carried out, followed by splitting the data into training and testing
sets to ensure reliable model evaluation. Advanced deep learning models, such as you only look once
(YOLO), fast region-based convolutional neural networks (Fast RCNN), ANNs, and LSTM networks, are
employed to capture complex patterns within the data [27]. These models are trained on the training set and
assessed on the testing set, using metrics like accuracy, precision, and recall to gauge their performance. A
comparative analysis is performed against traditional machine learning models, including decision trees and
SVMs, to highlight the superior predictive accuracy of deep learning techniques. This methodology aims to
provide precise insights into student performance, enabling more effective and targeted educational
interventions.

Figure 1. Proposed educational data mining models

3.1. SVM method used for EDM with spatial pyramid pooling
SVM is a powerful machine learning algorithm used for both classification and regression tasks.
SVM is often applied to predict student performance by classifying students into different performance
categories or predicting continuous scores [28]. Given a dataset of student features (such as grades,
attendance, and demographic data), the goal is to classify students into categories like pass/fail,
high/medium/low performance, or predict their final scores. The primary objective of SVM is to identify a
hyperplane that optimally separates data points (students) into distinct classes. In student performance
prediction, the hyperplane separates students based on their performance levels.

𝜔𝑇 𝑥 + 𝑏 = 0 (1)

𝑓(𝑥) = 𝜔𝑇 𝑥 + 𝑏 (2)

Educational data mining approach for predicting student performance and … (Muniappan Ramaraj)
4116  ISSN: 2252-8938

Where 𝜔 is the weight vector (which determines the orientation of the hyperplane), x is the feature vector
(input data, such as student features), b is the bias (offset from the origin), 𝜔𝑇 𝑥 + 𝑏 = 0 defines the
hyperplane, and this is the decision boundary. If f(x)≥0 the student is classified into one category (e.g., pass),
If f(x)<0 the student is classified into the other category (e.g., fail).

3.2. YOLO method used for EDM with spatial pyramid pooling
EDM, to the direct application of YOLO for student performance prediction is unconventional, as
YOLO is fundamentally designed for image-based tasks. However, with some creative modification, YOLO-
like architectures could theoretically be adapted for EDM tasks, especially if image-like data representations
(e.g., heatmaps, time series, or visual patterns of student activity) are used [29]. n traditional YOLO, the
objective is to predict bounding boxes around objects in an image and classify them. T The algorithm
segments the image into an S×S grid, where each grid cell predicts multiple bounding boxes along with
corresponding confidence scores and class probabilities.
𝑝
𝑆𝑃𝑃 = 𝜆𝑝 ∑𝑛𝑖=1(𝑃(𝑇𝑃𝑖 ) − 𝑃(𝑃𝑃𝑖 ))2 + 𝜆𝑐 ∑𝑛𝑖=1(𝑇𝐶𝑖 − 𝑃𝐶𝑖 )2 + 𝜆𝑓 ∑𝑛𝑖=1 ∑𝑚 𝑡
𝑗=1(𝑓𝑖,𝑗 − 𝑓𝑖,𝑗 )
2
(3)

Where 𝜆𝑝 , 𝜆𝑐 , 𝜆𝑓 are hyperparameters that control the relative importance of each loss term, (𝑃(𝑇𝑃𝑖 ) − 𝑃(𝑃𝑃𝑖 ))
𝑡 𝑝
are the true and predicted probabilities of student iii belonging to a certain performance class, (𝑓𝑖,𝑗 − 𝑓𝑖,𝑗 ) are the
true and predicted feature values for student i and feature j.

3.3. Fast RCNN method used for EDM with spatial pyramid pooling
Fast RCNN is a computer vision algorithm typically used for object detection in images. While it's
not directly applicable to EDM, it can creatively adapt the principles of Fast RCNN for student performance
prediction. The idea is to leverage its underlying framework for analyzing segmented data regions and
making predictions, and map these concepts onto the features and performance prediction tasks in EDM.

𝐿𝑐 = − ∑𝑛𝑖=1 ∑𝑚
𝑗=1 𝑦𝑖𝑗 log 𝑃 (𝑃𝑖𝑗 ) (4)

Where 𝑦𝑖𝑗 is the true performance class for student i’s region j, log 𝑃 (𝑃𝑖𝑗 ) is the predicted probability of the
true class for region j. For each student, we make predictions for each region of features and then aggregate
these to make a final decision about the student's overall performance. The algorithm can classify
performance or predict scores for each feature set and then aggregate these predictions to make a final
decision on student performance.

3.4. ANN with LSTM method used for EDM with spatial pyramid pooling
ANNs combined with LSTM units are widely used for time series prediction and sequential data
modeling. In EDM, this combination can be highly effective for student performance prediction, especially
when there is a temporal aspect to the data (e.g., predicting performance over multiple semesters or
assessments). A neural network with fully connected layers, typically used for learning from non-sequential,
static data. In EDM, an ANN can be used to model relationships between student features (e.g., grades,
attendance, and assignment scores) and their final performance. LSTMs are particularly useful in modeling
time-dependent relationships, such as a student’s performance over multiple periods or tasks.

𝐿𝑐 = − ∑𝑛𝑖=1 𝑦𝑖 log 𝑃(𝑃𝑖 ) (5)

𝐿𝑟 = 1⁄𝑛 ∑𝑛𝑖=1(𝑦𝑖 − 𝑦̂)

𝑖
2
(6)

Where 𝑦𝑖 is the true performance class for student i, log 𝑃(𝑃𝑖 ) is the predicted probability for the true class.
Then, next equation 𝑦𝑖 is the true performance class for student I, 𝑦̂𝑖 is the predicted performance score. For
each student, the LSTM processes the sequential features, and the ANN layers make the final performance
prediction based on the learned representation.

4. RESULTS AND DISCUSSION

The ANN and LSTM models demonstrated superior performance in predicting student outcomes
compared to traditional machine learning methods. The LSTM, in particular, excelled at capturing temporal
patterns in the data, leading to higher accuracy in predicting long-term student performance. When compared
to decision trees and SVMs, deep learning models showed a marked improvement in prediction accuracy.
The ANN and LSTM models reduced the error rate by approximately 10-15%, highlighting their
effectiveness in identifying non-linear relationships and complex patterns in student data.

Int J Artif Intell, Vol. 14, No. 5, October 2025: 4113-4122

Int J Artif Intell ISSN: 2252-8938  4117

4.1. Classification accuracy

The classification accuracy for student performance prediction in the real time dataset, to find the
accuracy is a commonly used metric to evaluate the performance of a classification model. Accuracy
measures the proportion of correct predictions made by the model relative to the total number of predictions.

𝐴 = 1⁄𝑛 ∑𝑛𝑖=1 1 (𝑦̂𝑖 = 𝑦𝑖 ) (7)

Where n is the total number of students, 𝑦̂𝑖 is predicted class either 0 or 1, 𝑦𝑖 is the true class of i.
In the context of student performance prediction, accuracy measures how well the model classifies
students into the correct performance categories (e.g., pass/fail, high/medium/low performance). Let’s say the
model is predicting whether a student will pass or fail based on their features (such as grades, attendance, and
assignments). If the model classifies a student as passing, and the student actually passes, it is a true positive
(TP). If it predicts failure, and the student fails, it is a true negative (TN).

4.2. Precision, recall, and F-measures

Precision calculates the proportion of TP predictions among all positive predictions (including both
TP and false positives (FP)). It addresses the question: "Out of all the students predicted to succeed, how
many actually did?" Recall, also referred to as sensitivity or the true positive rate (TPR), measures the ratio
of TP predictions to all actual positives (TP and false negatives (FN)). It answers: "Out of all the students
who actually succeeded, how many were correctly predicted by the model?" The F1-score, which is the
harmonic mean of precision and recall, offers a single metric that balances the two. This score is particularly
valuable when there is a need to balance precision and recall, such as when both FP and FN have significant
consequences.
𝑇𝑃
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = (8)
𝑇𝑃+𝐹𝑃

𝑇𝑃
𝑅𝑒𝑐𝑎𝑙𝑙 = (9)
𝑇𝑃+𝐹𝑁

𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛×𝑅𝑒𝑐𝑎𝑙𝑙
𝐹 − 𝑚𝑒𝑎𝑠𝑢𝑟𝑒 = 2 × (10)
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛+𝑅𝑒𝑐𝑎𝑙𝑙

Where TP refer to the number of cases where positive outcomes are correctly predicted (e.g., students
correctly identified as passing). FP represent instances where the model incorrectly predicts a positive
outcome (e.g., students predicted to pass but actually fail). FN are the cases where the model wrongly
predicts a negative outcome (e.g., students predicted to fail but actually pass). The F1-score ranges between
0 and 1, with 1 signifying perfect precision and recall.

4.3. Receiver operating characteristic

The receiver operating characteristic (ROC) curve is an effective tool for assessing the performance
of a classification model, especially when predicting student outcomes (such as pass/fail or different
performance categories). It is a graphical representation that illustrates the trade-off between the TPR and the
false positive rate (FPR) as the decision threshold changes. The ROC curve plots the TPR against the FPR for
varying thresholds, with each point on the curve reflecting a different threshold used by the model to classify
students as passing or failing. The x-axis represents the FPR, while the y-axis represents the TPR.
𝑇𝑃 𝐹𝑃
𝑇𝑃𝑅 = , 𝐹𝑃𝑅 = (11)
𝑇𝑃+𝐹𝑁 𝐹𝑃+𝑇𝑁

This represents one point on the ROC curve. By varying the threshold, you generate additional TPR and FPR
values to plot the entire curve.

4.4. Time calculation

EDM for student performance prediction, calculating the time complexity of the algorithms and the
overall prediction process can provide insights into the efficiency of the model. Time complexity generally
refers to the amount of computational time an algorithm takes as a function of the length of the input.

𝑜(𝑛. 𝑝2 ) (12)

Where n is the number of instances and p is the number of features. The computation involves calculating the
coefficients using the least squares method.

Educational data mining approach for predicting student performance and … (Muniappan Ramaraj)
4118  ISSN: 2252-8938

Table 1 represents that the comparison of various methods for student performance calculation
reveals distinct differences in accuracy and processing time. Among the techniques evaluated SVM, Fast
RCNN, ANN with LSTM, and YOLO. YOLO demonstrates the highest accuracy, achieving 0.86 during
training and 0.93 during testing, while also exhibiting the shortest processing time of 2.7 seconds for training
and 1.84 seconds for testing. Fast RCNN follows closely, with training and testing accuracies of 0.76 and
0.87, respectively, taking slightly longer at 3.7 seconds for training. ANN with LSTM shows competitive
performance with accuracies of 0.74 for training and 0.87 for testing, though it requires more time than
YOLO at 3.1 seconds for training. SVM, while effective, records the lowest accuracy of 0.73 during training
and 0.82 during testing, taking 3.4 seconds for training. Overall, YOLO stands out as the most efficient and
accurate method in this comparison.
In the Figure 2, the performance comparison of various methods like SVM, Fast RCNN, ANN with
LSTM, and YOLO illustrates that YOLO achieves the highest accuracy of 0.93 during testing while also
being the fastest, requiring only 1.84 seconds. Fast RCNN and ANN with LSTM provide competitive
accuracies of 0.87 but take longer, with Fast RCNN at 2.7 seconds and ANN with LSTM at 2.02 seconds. In
contrast, SVM shows the lowest accuracy of 0.82 during testing and a training time of 3.4 seconds. Overall,
the data indicate that YOLO is the most effective method in terms of both accuracy and processing time.

Table 1. Comparison of student performance data for testing and training process
Methods Training Testing
Accuracy Time Accuracy Time
SVM 0.73 3.4 0.82 2.9
Fast RCNN 0.76 3.7 0.87 2.7
ANN with LSTM 0.74 3.1 0.87 2.02
YOLO 0.86 2.7 0.93 1.84

4 SVM 3.4 3.7 3.1

Fast RCNN 2.7 2.9 2.7
3
Percentage

ANN with LSTM

YOLO 2.02 1.84
2
0.73 0.76 0.74 0.86 0.82 0.87 0.87 0.93
1
0
Accuracy Time Accuracy Time
Training Testing
Measures

Figure 2. Performance comparison of proposed methods

Table 2 presents a comparative analysis of various methods for evaluating student performance,
specifically focusing on precision, recall, and F-measure during training and testing phases. Among the
methods assessed, Fast RCNN demonstrates the highest precision (0.752) and F-measure (0.721) during
training, indicating its superior ability to identify relevant instances. The ANN with LSTM closely follows,
with a training precision of 0.748 and an F-measure of 0.723, showcasing its effectiveness in handling
sequential data. SVM yields respectable results, with a training precision of 0.735, while YOLO shows the
lowest performance across metrics, particularly with a training precision of 0.706. During testing,
Fast RCNN again leads with a precision of 0.725, followed by ANN with LSTM at 0.723. Overall,
Fast RCNN and ANN with LSTM exhibit consistent performance, suggesting their potential suitability for
applications requiring reliable performance metrics in educational contexts.
In Figure 3 compares the performance metrics precision, recall, and F-measure of various methods
used to evaluate student performance. The methods analyzed include SVM, Fast RCNN, ANN with LSTM,
and YOLO, with results presented for both training and testing phases. This analysis highlights the
effectiveness of each method in accurately assessing student performance metrics.

Table 2. Overall performance comparison of the proposed measures

Methods Training Testing
Precision Recall F-measure Precision Recall F-measure
SVM 0.735 0.735 0.704 0.693 0.697 0.687
Fast RCNN 0.752 0.746 0.721 0.725 0.734 0.712
ANN with LSTM 0.748 0.746 0.723 0.723 0.734 0.732
YOLO 0.706 0.713 0.692 0.712 0.701 0.685

Int J Artif Intell, Vol. 14, No. 5, October 2025: 4113-4122

Int J Artif Intell ISSN: 2252-8938  4119

0.76 0.752 0.748

0.746 0.746 0.734
0.735 0.735 0.723 0.734 0.732
0.74 0.723
0.721 0.725
0.713 0.712 0.712 SVM
0.72 0.706
Percentage

0.704 0.701 Fast RCNN

0.697
0.7 0.692 0.693
0.687 0.685 ANN with LSTM
0.68 YOLO

0.66

0.64
Precision Recall F-Measure Precision Recall F-Measure
Training Testing
Measures

Figure 3. Overall performance comparison of the different measures

The performance of various methods for student performance prediction is illustrated on the Table 3
through their ROC values, which indicate the models' effectiveness in distinguishing between classes.
Among the evaluated techniques, the SVM achieved the highest training ROC value of 0.804 and a testing
ROC of 0.81, suggesting a robust model. Fast RCNN follows with a notable training ROC of 0.863;
however, its testing ROC drops to 0.746, indicating potential overfitting. The ANN with LSTM achieved a
training ROC of 0.826 but also faced a decline in testing performance at 0.717. Finally, the YOLO method
demonstrated the lowest performance overall, with training and testing ROC values of 0.713 and 0.693,
respectively. This comparison highlights the varying effectiveness of these models, emphasizing the need for
further optimization, particularly for those with lower testing ROC scores.
Figure 4 illustrates the performance comparison of various methods using ROC values derived from
a real-time dataset for student performance prediction. The evaluated techniques include SVM, Fast RCNN,
ANN with LSTM, and YOLO. Among these, SVM stands out with the highest training ROC of 0.804 and a
testing ROC of 0.81, indicating its reliability. Fast RCNN has a strong training ROC of 0.863, but its testing
ROC declines to 0.746, suggesting potential overfitting. The ANN with LSTM achieves a training ROC of
0.826, with a testing ROC of 0.717, reflecting a similar trend. Conversely, the YOLO method shows the
lowest performance, with training and testing ROC values of 0.713 and 0.693, respectively. This comparison
underscores the varied effectiveness of these models and highlights the need for optimization, especially for
those with lower testing ROC scores.

Table 3. Performance comparison of ROC applied on student performance dataset

Methods Training Testing
ROC ROC
SVM 0.804 0.81
Fast RCNN 0.863 0.746
ANN with LSTM 0.826 0.717
YOLO 0.713 0.693

1
0.8
Accuracy

0.6
0.4
Training ROC
0.2 Testing ROC
0
SVM Fast RCNN ANN with LSTM YOLO
Measures

Figure 4. Performance comparison of ROC in real time dataset

5. CONCLUSION
This study demonstrates the transformative potential of deep learning in the realm EDM. In this
article has proposed many advanced deep learning techniques introduced such as YOLO, Fast RCNN, ANNs,
and LSTM networks, the research reveals that deep learning models significantly enhance the accuracy of
Educational data mining approach for predicting student performance and … (Muniappan Ramaraj)
4120  ISSN: 2252-8938

student performance predictions compared to traditional machine learning methods. The deep learning
approaches employed effectively capture intricate, non-linear relationships in diverse data sources, including
academic assessments, demographic information, and student behaviors. The comparative analysis shows
that these models outperform conventional techniques like decision trees and SVMs in terms of predictive
accuracy. The findings underscore that deep learning can offer a more nuanced understanding of student
performance and behavior, which is crucial for identifying at-risk students and implementing timely,
personalized interventions. This capability allows educational institutions to better tailor their support
strategies and improves overall student success. By integrating deep learning into educational practices,
institutions can move beyond one-size-fits-all solutions and develop more effective, individualized
approaches to learning. This study highlights the significant potential of deep learning to revolutionize
personalized education, offering deeper insights into student needs and enhancing educational outcomes.

FUNDING INFORMATION
Authors state no funding involved.

AUTHOR CONTRIBUTIONS STATEMENT

This journal uses the Contributor Roles Taxonomy (CRediT) to recognize individual author
contributions, reduce authorship disputes, and facilitate collaboration.

Name of Author C M So Va Fo I R D O E Vi Su P Fu
Muniappan Ramaraj ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Sabareeswaran ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Dhendapani
Jothish Chembath ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Selvaraj Srividhya ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Nainan Thangarasu ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Bhaarathi Ilango ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

C : Conceptualization I : Investigation Vi : Visualization

M : Methodology R : Resources Su : Supervision
So : Software D : Data Curation P : Project administration
Va : Validation O : Writing - Original Draft Fu : Funding acquisition
Fo : Formal analysis E : Writing - Review & Editing

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no conflicts of interest related to this work.

INFORMED CONSENT
We have obtained informed consent from all participants involved in the study.

ETHICAL APPROVAL
The authors have obtained the necessary permissions from the institutional ethics committee to
conduct this work.

DATA AVAILABILITY
Data availability is not applicable to this paper as no new data were created or analyzed in this study.

REFERENCES
[1] X. Liu, “The educational resource management based on image data visualization and deep learning,” Heliyon, vol. 10, no. 13,
2024, doi: 10.1016/j.heliyon.2024.e32972.
[2] B. Alnasyan, M. Basheri, and M. Alassafi, “The power of deep learning techniques for predicting student performance in virtual
learning environments: a systematic literature review,” Computers and Education: Artificial Intelligence, vol. 6, 2024,
doi: 10.1016/j.caeai.2024.100231.
[3] R. Cerezo, J.-A. Lara, R. Azevedo, and C. Romero, “Reviewing the differences between learning analytics and educational data
mining: towards educational data science,” Computers in Human Behavior, vol. 154, 2024, doi: 10.1016/j.chb.2024.108155.
[4] L. Gao, Z. Zhao, C. Li, J. Zhao, and Q. Zeng, “Deep cognitive diagnosis model for predicting students’ performance,” Future
Generation Computer Systems, vol. 126, pp. 252–262, 2022, doi: 10.1016/j.future.2021.08.019.

Int J Artif Intell, Vol. 14, No. 5, October 2025: 4113-4122

Int J Artif Intell ISSN: 2252-8938  4121

[5] X. Song, J. Li, T. Cai, S. Yang, T. Yang, and C. Liu, “A survey on deep learning based knowledge tracing,” Knowledge-Based
Systems, vol. 258, 2022, doi: 10.1016/j.knosys.2022.110036.
[6] A. Bressane et al., “Understanding the role of study strategies and learning disabilities on student academic performance to
enhance educational approaches: a proposal using artificial intelligence,” Computers and Education: Artificial Intelligence, vol. 6,
2024, doi: 10.1016/j.caeai.2023.100196.
[7] K. Okoye, J. T. Nganji, J. Escamilla, and S. Hosseini, “Machine learning model (RG-DMML) and ensemble algorithm for
prediction of students’ retention and graduation in education,” Computers and Education: Artificial Intelligence, vol. 6, 2024,
doi: 10.1016/j.caeai.2024.100205.
[8] S. Maniyan, R. Ghousi, and A. Haeri, “Data mining-based decision support system for educational decision makers: Extracting
rules to enhance academic efficiency,” Computers and Education: Artificial Intelligence, vol. 6, 2024,
doi: 10.1016/j.caeai.2024.100242.
[9] A. Harb et al., “Diverse distant-students deep emotion recognition and visualization,” Computers and Electrical Engineering,
vol. 111, 2023, doi: 10.1016/j.compeleceng.2023.108963.
[10] T. Shaik, X. Tao, C. Dann, H. Xie, Y. Li, and L. Galligan, “Sentiment analysis and opinion mining on educational data: a survey,”
Natural Language Processing Journal, vol. 2, 2023, doi: 10.1016/j.nlp.2022.100003.
[11] S. Rathi, K. K. Hiran, and S. Sakhare, “Affective state prediction of E-learner using SS-ROA based deep LSTM,” Array, vol. 19,
2023, doi: 10.1016/j.array.2023.100315.
[12] J. Ding, “Deep learning perspective on the construction of SPOC teaching model of music and dance in colleges and universities,”
Systems and Soft Computing, vol. 6, 2024, doi: 10.1016/j.sasc.2024.200137.
[13] K. Aulakh, R. K. Roul, and M. Kaushal, “E-learning enhancement through educational data mining with COVID-19 outbreak
period in backdrop: a review,” International Journal of Educational Development, vol. 101, 2023,
doi: 10.1016/j.ijedudev.2023.102814.
[14] S. Sarker, M. K. Paul, S. T. H. Thasin, and M. A. M. Hasan, “Analyzing students’ academic performance using educational data
mining,” Computers and Education: Artificial Intelligence, vol. 7, 2024, doi: 10.1016/j.caeai.2024.100263.
[15] G. Feng and M. Fan, “Research on learning behavior patterns from the perspective of educational data mining: evaluation,
prediction and visualization,” Expert Systems with Applications, vol. 237, 2024, doi: 10.1016/j.eswa.2023.121555.
[16] J. Deng, X. Huang, and X. Ren, “A multidimensional analysis of self-esteem and individualism: A deep learning-based model for predicting
elementary school students’ academic performance,” Measurement: Sensors, vol. 33, 2024, doi: 10.1016/j.measen.2024.101147.
[17] P. X. Lam, P. Q. H. Mai, Q. H. Nguyen, T. Pham, T. H. H. Nguyen, and T. H. Nguyen, “Enhancing educational evaluation
through predictive student assessment modeling,” Computers and Education: Artificial Intelligence, vol. 6, 2024,
doi: 10.1016/j.caeai.2024.100244.
[18] J. Peng et al., “DeepRisk: a deep learning approach for genome-wide assessment of common disease risk,” Fundamental
Research, vol. 4, no. 4, pp. 752–760, 2024, doi: 10.1016/j.fmre.2024.02.015.
[19] A. Rejeb, K. Rejeb, A. Appolloni, H. Treiblmaier, and M. Iranmanesh, “Exploring the impact of ChatGPT on education: a web mining
and machine learning approach,” The International Journal of Management Education, vol. 22, 2024, doi: 10.1016/j.ijme.2024.100932.
[20] P. Bhardwaj, P. K. Gupta, H. Panwar, M. K. Siddiqui, R. M.-Menendez, and A. Bhaik, “Application of deep learning on student
engagement in e-learning environments,” Computers & Electrical Engineering, vol. 93, 2021, doi: 10.1016/j.compeleceng.2021.107277.
[21] A. Al Ka’bi, “Proposed artificial intelligence algorithm and deep learning techniques for development of higher education,”
International Journal of Intelligent Networks, vol. 4, pp. 68–73, 2023, doi: 10.1016/j.ijin.2023.03.002.
[22] C.-C. Lin, E. S. J. Cheng, A. Y. Q. Huang, and S. J. H. Yang, “DNA of learning behaviors: A novel approach of learning performance
prediction by NLP,” Computers and Education: Artificial Intelligence, vol. 6, 2024, doi: 10.1016/j.caeai.2024.100227.
[23] H. Farhood, I. Joudah, A. Beheshti, and S. Muller, “Advancing student outcome predictions through generative adversarial
networks,” Computers and Education: Artificial Intelligence, vol. 7, 2024, doi: 10.1016/j.caeai.2024.100293.
[24] S. Riaz, A. Saghir, M. J. Khan, H. Khan, H. S. Khan, and M. J. Khan, “TransLSTM: a hybrid LSTM-Transformer model for fine-
grained suggestion mining,” Natural Language Processing Journal, vol. 8, Sep. 2024, doi: 10.1016/j.nlp.2024.100089.
[25] Y. Wang, Y. Zhang, M. Liang, R. Yuan, J. Feng, and J. Wu, “National student loans default risk prediction: A heterogeneous
ensemble learning approach and the SHAP method,” Computers and Education: Artificial Intelligence, vol. 5, 2023,
doi: 10.1016/j.caeai.2023.100166.
[26] R. Song, F. Pang, H. Jiang, and H. Zhu, “A machine learning based method for constructing group profiles of university
students,” Heliyon, vol. 10, no. 7, 2024, doi: 10.1016/j.heliyon.2024.e29181.
[27] M. Bilal, H. Israr, M. Shahid, and A. Khan, “Sentiment classification of roman-urdu opinions using naïve Bayesian, decision tree
and KNN classification techniques,” Journal of King Saud University - Computer and Information Sciences, vol. 28, no. 3,
pp. 330–344, 2016, doi: 10.1016/j.jksuci.2015.11.003.
[28] X. Guo et al., “Longitudinal machine learning prediction of non-suicidal self-injury among Chinese adolescents: a prospective
multicenter Cohort study,” Journal of Affective Disorders, vol. 392, 2026, doi: 10.1016/j.jad.2025.120110.
[29] S. Gupta and J. Choudhary, “An efficient test suit reduction methodology for regression testing,” Indonesian Journal of Electrical
Engineering and Computer Science, vol. 34, no. 2, pp. 1336–1343, 2024, doi: 10.11591/ijeecs.v34.i2.pp1336-1343.

BIOGRAPHIES OF AUTHORS

Muniappan Ramaraj is working as an Assistant Professor in the Department of

Computer Science at Rathinam College of Arts and Science, Coimbatore. He holds a Ph.D.,
degree in Computer Science at Bharathiar University in the year of 2020 with specialization in
data mining with image process and also fuzzy logic in the image analysis. His research areas
are data mining, image processing, fuzzy logic, pattern recognition, and deep learning concept.
He has published more research article in the reputed various national and international
journals and also filed the patents in the same field. He has a reviewer of many international
journals including with IEEE, ASTESJ, and JERS. He can be contacted at email:
[email protected].

Educational data mining approach for predicting student performance and … (Muniappan Ramaraj)
4122  ISSN: 2252-8938

Dr. Sabareeswaran Dhendapani is working as an Assistant Professor in the

Department of Computer Science at L.R.G. Government College of Arts and Science for
Women, Tiruppur. He holds a Ph.D., degree in computer science at Karpagam Academy of
Higher Education in the year of 2020 with specialization in data mining. His research areas are
data mining, image processing, fuzzy logic, and pattern recognition. He has published more
research article in the reputed various national and international journals and also filed the
patents in the same field. He can be contacted at email: [email protected].

Dr. Jothish Chembath completed his Ph.D. from Karpagam deemed to be

University, Coimbatore, Tamilnadu in the year 2020, under the title “Next web page
prediction using enhanced preprocessing and ensemble clustering based hybrid Markov model
using web log data”. More than 15 research journals have been published in reputed journals
which are either Scopus indexed, Web of Science, and international journals of repute, which
would help the Internet mechanism to predict the user’s intention and interest, when internet is
browsed using mathematical model algorithms after cleaning the internet data, thereby giving
accurate predictions using mathematical models. He also has attended 3 international
conferences held in the Sultanate of Oman, Malaysia, and Cochin. Currently author is working
in Presidency University, Bengaluru. The teaching and research experience is around two
decades starting from 2003 to toll date. He can be contacted at email:
[email protected].

Selvaraj Srividhya is working as an Associate Professor in the Department of

Computer Science at KPR College of Arts Science a Research, Coimbatore. She has more
than a decade of teaching experience in different verticals. She is a reviewer in various
international journals and life-time member in IAENG. Her research area includes data
mining, machine learning, and deep learning. She has about 10+ research papers published in
national and international journals. She also presented more than 15 papers in national and
international conferences and also published many books. She can be contacted at email:
[email protected].

Dr. Nainan Thangarasu is currently working as an Assistant Professor in the

Department of Computer Science, at Karpagam Academy of Higher Education, Coimbatore.
He is greatly fascinated with the advanced computing technology and research programs is
cluster computing, cryptography and network security, cloud computing, artificial intelligent
system, information security in large database and data mining as well as the strong teaching
experience. His doctoral dissertation also focuses on advanced security systems with cloud
computing, and he has published more than 13 publications in reputed journals, which he finds
would be a great addition to the success of your teaching and research department. He can be
contacted at email: [email protected].

Bhaarathi Ilango is working as an Assistant Professor in the Department of

Computer Science at Rathinam College of Arts and Science, Coimbatore. He holds a NET.
Qualification in Computer Science at NTA in the year of 2023. with specialization in data
mining with image process and also fuzzy logic in the image analysis. His research areas are
data mining, in the various fields. He has published more research article in the reputed
various national and international journals and also filed the patents in the same field. He can
be contacted at email: [email protected].

Int J Artif Intell, Vol. 14, No. 5, October 2025: 4113-4122