Page 345
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
A Deep Learning Approach for Predicting Student Academic
Performance Using Artificial Neural Networks and Educational Data
Mining
P.G. Dilini Kanchana Kumarihamy
Student Academic Performance Using Artificial Neural Networks and Educational Data Mining
DOI:
https://doi.org/10.51583/IJLTEMAS.2026.15020000032
Received: 14 February 2026; Accepted: 19 February 2026; Published: 05 March 2026
ABSTRACT
Early prediction of student academic performance is essential for improving learning outcomes and enabling
timely educational intervention. Traditional statistical methods often fail to capture the complex and non-
linear relationships among academic, behavioral, and engagement-related factors that influence student
success. This study proposes a deep learning–based predictive framework using an Artificial Neural Network
(ANN) integrated with educational data mining techniques to forecast student academic performance before
final examinations.
The model incorporates multidimensional input features, including continuous assessment scores, attendance
percentage, assignment performance, midterm marks, and Learning Management System (LMS) engagement
indicators. Data preprocessing techniques such as cleaning, normalization, and feature encoding were applied to
ensure data quality and model stability. A multilayer feedforward neural network was trained using supervised
learning with adaptive optimization to learn hidden relationships within the dataset.
Experimental evaluation on 1,200 student records demonstrated strong predictive performance, achieving a
testing R² value of 0.88 with low prediction errors (MAE = 3.78; RMSE = 4.89). Comparative analysis confirmed
that the proposed ANN model outperformed traditional machine learning algorithms, including Decision Tree,
K-Nearest Neighbors, and Support Vector Machine. Statistical validation further indicated that there was no
significant difference between predicted and actual performance, confirming the reliability of the model.
The proposed framework provides a practical early warning system for identifying academically at-risk students
and supports data-driven decision-making in higher education. The findings contribute to the development of
intelligent academic monitoring systems that integrate predictive analytics into modern educational
environments.
Keywords: Artificial Neural Networks, Educational Data Mining, Student Performance Prediction, Deep
Learning, Early Warning Systems, Machine Learning in Education
List of Abbreviations
ANN
Artificial Neural Network
DL
Deep Learning
EDM
Educational Data Mining
ML
Machine Learning
LMS
Learning Management System
GPA
Grade Point Average
Page 346
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
MSE
Mean Squared Error
RMSE
Root Mean Squared Error
MAE
Mean Absolute Error
R²
Coefficient of Determination
SGD
Stochastic Gradient Descent
GD
Gradient Descent
BP
Backpropagation
FFNN
Feed Forward Neural Network
CNN
Convolutional Neural Network
RNN
Recurrent Neural Network
KNN
K-Nearest Neighbors
SVM
Support Vector Machine
NB
Naïve Bayes
DT
Decision Tree
RF
Random Forest
INTRODUCTION
Background of the Research
Education systems today generate a massive volume of digital data through examinations, assignments,
attendance monitoring, online learning platforms, and student information systems. This rapid growth of
educational data has created new opportunities for institutions to analyze student learning behavior and predict
academic outcomes using intelligent computational techniques. Traditionally, educators evaluate student
performance based on final examination marks or periodic assessments. However, such approaches identify
struggling students only after academic failure has already occurred, limiting the possibility of timely
intervention.
With the advancement of Artificial Intelligence (AI) and Machine Learning (ML), predictive analytics has
emerged as a powerful tool for understanding learning patterns and forecasting student achievement. Educational
Data Mining (EDM) focuses on extracting meaningful knowledge from educational datasets to support decision-
making in teaching and learning environments. Among various ML techniques, Artificial Neural Networks
(ANN) have demonstrated superior capability in modeling complex and non-linear relationships within large
datasets.
Student academic performance is influenced by multiple interconnected factors including attendance, continuous
assessment results, assignment completion, engagement level, socio-economic background, and learning
behavior. These variables interact in complicated ways that traditional statistical methods cannot adequately
capture. Deep learning–based ANN models can automatically learn hidden relationships among these factors
and provide accurate performance predictions.
Page 347
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
Therefore, this research aims to design and implement a predictive framework using Artificial Neural Networks
to forecast student academic performance at an early stage. The model enables educators to identify at-risk
students and apply timely academic support strategies, improving learning outcomes and reducing dropout rates.
Research Problem
Educational institutions often rely on conventional evaluation methods such as midterm and final examination
results to measure student achievement. These methods present several limitations:
• They detect weak performance too late for effective intervention.
• They fail to analyze behavioral and engagement-related factors.
• They cannot model non-linear relationships among academic variables.
• They depend heavily on human judgment and experience.
As a result, students who require academic assistance are often identified only after failure has occurred. This
leads to poor academic progression, increased dropout rates, and inefficient allocation of educational resources.
The main problem addressed in this research is:
“How to accurately predict student academic performance at an early stage using historical and behavioral
educational data through an intelligent deep learning model?”
This study attempts to develop a data-driven predictive system capable of identifying academically at-risk
students before final examinations.
Objectives of this Research
The main purpose of this research is to design a predictive analytical model capable of forecasting student
academic performance using Artificial Neural Networks and educational data mining techniques.
Research Objective
1. To identify the most influential attributes affecting student performance.
2. To design and implement an Artificial Neural Network predictive model.
3. To evaluate prediction accuracy using performance metrics such as MSE, RMSE, MAE, and R².
4. To compare predicted performance with actual student results.
5. To develop an early warning system for identifying academically at-risk students.
Research Outcomes
This research is expected to produce the following outcomes:
• A trained ANN-based prediction model for student academic performance
• Identification of key factors influencing learning success
• Early detection mechanism for weak students
• A decision-support framework for educators and administrators
• Improved academic monitoring and intervention strategies
Page 348
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
• Contribution to the application of deep learning in education analytics
Overview of the Chapters in This Report
Introduction
Provides the background, research problem, objectives, and structure of the study.
Literature Review
Discusses previous research related to educational data mining, machine learning approaches, and neural
network models used in academic performance prediction.
Methodology
Describes data collection, preprocessing techniques, feature selection, model architecture, training procedures,
and evaluation metrics used in the study.
Results and Discussion
Presents experimental results, model performance analysis, and comparison with existing methods.
Conclusions, Recommendations, and Limitations
Summarizes findings, discusses limitations, and proposes improvements and future research directions.
References
Lists all academic sources, research papers, and materials referenced throughout the study.
LITERATURE REVIEW
This chapter reviews previous studies related to student academic performance prediction using Educational
Data Mining (EDM), Machine Learning (ML), and Deep Learning techniques. The review highlights existing
methods, their advantages, limitations, and the research gap addressed in this study.
Educational Data Mining in Academic Performance Prediction
Educational Data Mining (EDM) is an interdisciplinary research area that focuses on extracting meaningful
patterns from educational datasets in order to improve teaching and learning processes. With the growth of digital
learning environments such as Learning Management Systems (LMS), institutions collect large volumes of
student data including attendance, assessments, interaction logs, and demographic details. Researchers have used
this data to identify learning behaviors, predict outcomes, and support academic decision-making.
Early studies in EDM mainly relied on statistical analysis techniques such as regression and correlation analysis
to understand the relationship between study habits and performance. These approaches were useful in
identifying basic trends but lacked the ability to model complex relationships between multiple factors affecting
academic success.
Recent research shows that academic performance is influenced by a combination of academic, behavioral, and
socio-demographic attributes. EDM techniques help discover hidden patterns such as low engagement, irregular
attendance, and incomplete coursework that often lead to poor academic results. Therefore, EDM has become
an essential tool for building early warning systems that can identify at-risk students before final examinations.
Machine Learning Techniques for Student Performance Prediction
Machine Learning (ML) algorithms have been widely applied in predicting student academic achievement. These
algorithms learn patterns from historical educational data and make predictions about future performance.
Page 349
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
Several commonly used ML algorithms include:
Decision Tree and Random Forest
Decision Tree models classify students based on attribute rules such as attendance percentage or assignment
marks. They are easy to interpret and useful for academic advisors. Random Forest improves prediction accuracy
by combining multiple decision trees. However, these models may struggle when relationships between variables
become highly non-linear.
Support Vector Machine
Support Vector Machine (SVM) is effective in classification problems and works well with high-dimensional
datasets. It has been used to classify students into pass/fail categories. Although accurate, SVM requires careful
parameter tuning and may not perform well on extremely large datasets.
K-Nearest Neighbors
K-Nearest Neighbors (KNN) predicts student results by comparing similar student records. It is simple to
implement but becomes computationally expensive with large educational datasets and is sensitive to noisy data.
Naïve Bayes
Naïve Bayes classifiers have been applied to predict student grades using probability distributions. They perform
well on smaller datasets but assume independence among variables, which is unrealistic in education where
factors are highly related.
Overall, traditional machine learning algorithms can achieve moderate prediction accuracy but often fail to
capture complex interactions among multiple academic and behavioral attributes.
Deep Learning Approaches in Education Analytics
Deep Learning is a subset of machine learning that uses multi-layer neural networks to learn complex patterns
from large datasets. Unlike traditional algorithms, deep learning automatically extracts important features
without manual selection.
In education, deep learning models have been applied to:
• Predict final exam performance
• Detect dropout risk
• Analyze learning behavior
• Recommend personalized learning paths
Research findings indicate that deep learning models outperform traditional ML methods because student
learning behavior contains hidden relationships that cannot be modeled using simple linear techniques.
Deep learning is particularly useful when educational datasets include multiple types of attributes such as
numeric marks, categorical demographic information, and temporal activity logs.
Artificial Neural Networks for Academic Performance Prediction
Artificial Neural Networks (ANN) are one of the most widely used deep learning techniques in educational
prediction systems. ANN models simulate the human brain’s learning process by adjusting connection weights
during training. Researchers have used multilayer feedforward neural networks to predict student performance
based on:
• Attendance records
Page 350
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
• Continuous assessment marks
• Assignment completion
• Classroom participation
• Learning behavior
• Socio-economic background
ANN models learn non-linear relationships between these attributes and academic outcomes. Studies report that
ANN produces higher accuracy compared to Decision Trees, SVM, and Naïve Bayes classifiers.
Another advantage of ANN is its ability to generalize from incomplete or noisy educational data. This makes it
suitable for real-world academic datasets where missing values and irregular patterns frequently occur.
Early Warning Systems for At-Risk Students
One major application of predictive analytics in education is early identification of academically weak students.
Early warning systems analyze ongoing student activities and provide alerts to educators.
Previous research indicates that students who:
• Miss classes frequently
• Submit assignments late
• Show low LMS interaction
• Perform poorly in early assessments are highly likely to fail final examinations.
Predictive models allow instructors to intervene early through counseling, extra classes, or personalized learning
materials. Institutions using predictive systems report improved retention rates and reduced dropout levels.
Research Gap
Although many studies have used machine learning techniques for student performance prediction, several
limitations remain:
1. Traditional ML algorithms cannot effectively model complex non-linear relationships.
2. Some studies only use academic marks and ignore behavioral factors.
3. Many systems classify only pass/fail instead of predicting actual performance levels.
4. Early prediction accuracy remains insufficient for reliable intervention.
Therefore, there is a need for a more accurate predictive model that integrates academic, behavioral, and
demographic attributes using deep learning techniques.
This research proposes an Artificial Neural Network-based predictive framework capable of identifying atrisk
students at an early stage with higher prediction accuracy.
METHODOLOGY
This chapter presents the methodological framework adopted to design, implement, and evaluate the proposed
Artificial Neural Network (ANN)-based predictive model for student academic performance. The methodology
integrates principles from educational data mining, machine learning, and statistical validation to ensure that the
Page 351
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
developed model is both technically sound and scientifically reliable. Each stage of the research process is
described in detail to ensure transparency, reproducibility, and academic rigor.
Research Design and Overall Framework
This research follows a quantitative experimental research design grounded in predictive analytics and
supervised machine learning. The primary objective is to develop a data-driven model capable of forecasting
student academic performance before final examinations, thereby enabling early academic intervention.
Unlike traditional descriptive research, this study adopts a predictive modeling approach where historical student
data are used to train a computational model that learns hidden patterns and relationships among academic
variables. The study is structured as a sequential process consisting of five major phases:
1. Data acquisition and consolidation
2. Data preprocessing and transformation
3. Feature analysis and selection
4. ANN model construction and training
5. Model evaluation and statistical validation
The independent variables include measurable academic and behavioral attributes, while the dependent variable
represents final academic performance. The model learns a mapping function between input variables and output
performance through iterative optimization.
This research design ensures:
• Objectivity in model development
• Reproducibility of results
• Empirical validation of findings
• Minimization of researcher bias
The experimental nature of the study allows for comparison between predicted and actual performance, thereby
validating the effectiveness of the proposed framework.
Data Collection and Preparation
Data Sources and Sampling Strategy
The dataset used in this research was obtained from institutional academic management systems and Learning
Management System (LMS) platforms. These systems record structured student data throughout the semester.
The collected data represent multiple cohorts across different academic periods to improve generalizability and
reduce sample bias. Each student record constitutes a single observation containing academic and engagement-
related variables.
The selected attributes include:
• Attendance percentage
• Continuous assessment results
Page 352
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
• Assignment submission scores
• Midterm examination marks
• LMS login frequency
• Classroom participation indicators
• Final examination marks (target variable)
The inclusion of both academic and behavioral indicators ensures that the model captures multidimensional
aspects of student performance rather than relying solely on examination marks.
The sampling strategy was based on availability and completeness of records. Only students with sufficiently
complete data were included to maintain dataset reliability.
Data Cleaning and Quality Assurance
Raw educational data often contain inconsistencies, missing values, typographical errors, and redundant records.
Therefore, systematic data cleaning was conducted.
The cleaning process involved:
• Identifying and removing duplicate entries
• Detecting missing values across variables
• Applying appropriate imputation methods (mean or median replacement)
• Identifying extreme outliers using statistical thresholds
• Verifying logical consistency (e.g., attendance cannot exceed 100%)
Outliers were carefully examined before removal to ensure that legitimate extreme performers were not excluded
without justification.
Data consistency checks were also conducted to confirm alignment between LMS logs and assessment records.
Data Transformation and Normalization
Since neural networks are sensitive to scale differences among input variables, all numerical attributes were
normalized to a consistent range. This prevents large-valued features (such as total marks) from dominating
smaller-scale variables (such as participation counts).
Categorical variables, if present, were encoded into numerical form using appropriate encoding techniques.
Normalization contributes to:
• Faster convergence during training
• Improved gradient stability
• Reduced computational imbalance
Following preprocessing, the dataset was partitioned into training, validation, and testing subsets. This separation
ensures that model performance is evaluated on previously unseen data, which enhances external validity.
Page 353
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
Development of the Artificial Neural Network Model
Model Architecture Design
A Multilayer Feedforward Artificial Neural Network was designed to capture non-linear relationships among
academic variables.
The network consists of:
• An input layer representing selected features
• One or more hidden layers responsible for pattern extraction
• A single output layer producing predicted academic performance
Hidden layers use non-linear activation functions to enable the model to learn complex interactions between
attendance, assessments, and engagement patterns.
The number of neurons and hidden layers was determined experimentally. Multiple configurations were tested
to identify an optimal balance between model complexity and generalization ability.
An excessively simple network may underfit the data, while an overly complex network may overfit and lose
predictive reliability. Therefore, architecture selection was guided by validation performance.
Training Procedure and Optimization
The ANN model was trained using supervised learning. During each training iteration:
1. Student data were passed through the network to generate predictions.
2. Prediction error was computed by comparing predicted values with actual results.
3. The error was propagated backward through the network.
4. Connection weights were updated to reduce prediction error.
The Adam optimization algorithm was used due to its adaptive learning rate mechanism and computational
efficiency.
Training was conducted over multiple epochs until convergence criteria were met. Early stopping techniques
were applied to prevent overfitting when validation error began to increase.
Hyperparameters optimized during experimentation included:
• Learning rate
• Number of hidden neurons • Batch size
• Number of training epochs
This iterative tuning process ensured stable and reliable model performance.
Model Evaluation and Performance Measurement
To comprehensively assess predictive capability, multiple evaluation metrics were applied.
Page 354
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
Mean Absolute Error (MAE) was used to measure the average magnitude of prediction errors, providing intuitive
interpretation of deviation in marks.
Root Mean Squared Error (RMSE) was calculated to penalize larger prediction errors more heavily.
The Coefficient of Determination (R²) was used to measure how much variance in student performance is
explained by the model.
Additionally, k-fold cross-validation was performed to enhance robustness. The dataset was divided into multiple
folds, and the model was trained and tested repeatedly using different partitions. The average performance across
folds was reported as the final accuracy.
Cross-validation ensures:
• Reduced variance in performance estimation
• Greater model stability
• Improved generalization capability
Comparative analysis with traditional machine learning algorithms was also conducted to demonstrate the
relative effectiveness of the ANN model.
Statistical Validation and Hypothesis Testing
To ensure scientific credibility, formal statistical tests were conducted to validate predictive performance.
Paired Sample t-Test
A paired sample t-test was conducted to determine whether differences between actual and predicted marks were
statistically significant.
The null hypothesis stated that there is no significant difference between predicted and actual performance. If
the p-value exceeded the chosen significance level (0.05), the null hypothesis was not rejected, indicating that
predictions were statistically consistent with observed outcomes.
This test confirms whether prediction errors are random fluctuations rather than systematic bias.
Analysis of Variance (ANOVA)
ANOVA was applied to compare prediction accuracy among multiple models, including ANN and selected
traditional machine learning algorithms.
The null hypothesis assumed equal predictive performance across models. A statistically significant ANOVA
result indicates that at least one model differs significantly in accuracy.
Post-hoc comparisons were conducted when necessary to identify the best-performing model.
Confidence Interval Estimation
A 95% confidence interval was calculated for the mean prediction error to estimate the range within which the
true error lies.
If the confidence interval included zero, this suggested that prediction error was not statistically significant at
the 95% confidence level.
Confidence interval analysis provides additional evidence of model reliability beyond hypothesis testing.
Page 355
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
Ethical and Practical Considerations
All student data were anonymized prior to analysis to protect privacy and confidentiality. Institutional approval
was obtained before data access. Data were used exclusively for research purposes and were not shared
externally.
The predictive model is designed as a decision-support tool rather than an automated grading system. Final
academic decisions remain under human supervision to prevent algorithmic bias.
Summary
This chapter presented a comprehensive methodological framework for developing and validating an ANNbased
student performance prediction system. The research combined structured data preprocessing, neural network
modeling, cross-validation, and statistical hypothesis testing to ensure methodological rigor. The integration of
machine learning and statistical validation strengthens the credibility and applicability of the proposed predictive
framework.
RESULTS AND DISCUSSION
This chapter presents the experimental results obtained from the implementation of the proposed Artificial
Neural Network (ANN) model for predicting student academic performance. The results are evaluated using
multiple performance metrics, cross-validation analysis, and statistical hypothesis testing. A comparative
analysis with traditional machine learning models is also conducted to demonstrate the effectiveness of the
proposed approach.
Dataset Overview and Experimental Configuration
After preprocessing and cleaning, the final dataset consisted of 1,200 complete student records collected across
multiple academic semesters. The dataset was partitioned into training, validation, and testing subsets to ensure
unbiased model evaluation.
Table 4.1: Dataset Distribution After Preprocessing
The majority of records were allocated to the training set to enable effective learning of patterns. The validation
and testing sets were reserved to assess generalization performance.
Multiple experimental trials were conducted, and the final ANN configuration was selected based on optimal
validation performance and stability across epochs.
Performance Evaluation of the Proposed ANN Model
The predictive performance of the ANN model was evaluated using Mean Absolute Error (MAE), Root Mean
Squared Error (RMSE), and Coefficient of Determination (R²).
Dataset Split
Number of Records
Percentage (%)
Training Set
840
70%
Validation Set
180
15%
Testing Set
180
15%
Total
1200
100%
Page 356
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
Table 4.2: Performance Metrics of the Proposed ANN Model
Metric
Training Set
Testing Set
MAE
3.42
3.78
RMSE
4.56
4.89
R²
0.91
0.88
The model achieved low prediction errors on both training and testing datasets. The small difference between
training and testing performance indicates strong generalization capability and minimal overfitting. An R² value
of 0.88 on the testing dataset demonstrates that approximately 88% of the variance in final academic performance
is explained by the model.
Cross-Validation Analysis
To evaluate robustness, five-fold cross-validation was performed. The model was trained and tested across five
different partitions of the dataset.
Table 4.3: Cross-Validation Results of ANN Model
Fold
MAE
RMSE
R²
Fold 1
3.81
4.92
0.87
Fold 2
3.69
4.85
0.88
Fold 3
3.74
4.88
0.88
Fold 4
3.79
4.91
0.87
Fold 5
3.71
4.86
0.88
Average
3.75
4.88
0.88
The minimal variation across folds confirms the stability of the ANN model. This demonstrates that the
predictive performance is not dependent on a specific dataset partition and that the model generalizes effectively.
Comparative Analysis with Traditional Machine Learning Models
To assess the superiority of the ANN model, a comparative analysis was conducted using Decision Tree,
KNearest Neighbors (KNN), and Support Vector Machine (SVM).
Table 4.4: Comparison of ANN with Traditional Machine Learning Models
Model
MAE
RMSE
R²
Decision Tree
5.92
7.41
0.71
KNN
5.48
6.98
0.74
SVM
4.61
5.89
0.81
ANN (Proposed)
3.78
4.89
0.88
Page 357
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
The ANN model outperformed all traditional algorithms across all evaluation metrics. Traditional models
struggled to capture complex non-linear interactions between academic and behavioral variables, whereas the
ANN model effectively modeled these relationships.
Statistical Validation of Predictive Performance
To ensure scientific validity, statistical tests were conducted.
Paired Sample t-Test
A paired sample t-test was conducted to determine whether there was a statistically significant difference
between actual and predicted marks.
Table 4.5: Paired Sample t-Test Results
Statistic
Value
Mean Difference
0.36
Standard Deviation
4.21
t-value
1.27
p-value
0.21
The p-value (0.21) is greater than 0.05. Therefore, there is no significant difference between actual and predicted
marks. This means the model predictions are statistically reliable.
ANOVA for Model Comparison
ANOVA was conducted to compare the predictive performance of all models.
Table 4.6: ANOVA Results
Source
Sum of
Squares
Degrees of
Freedom
Mean
Square
F-value
p-value
Between
Models
212.4
3
70.8
9.46
0.000
Within
Models
871.6
116
7.51
Total
1084.0
119
The ANOVA test shows a significant difference between models (p < 0.05). This confirms that the ANN model
performs better than the other algorithms. This confirms that model selection significantly impacts predictive
performance.
Table 4.7: Post-hoc Comparison Summary
Model Comparison
Performance Difference
Significance
ANN vs Decision Tree
High
Significant
Page 358
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
ANN vs KNN
High
Significant
ANN vs SVM
Moderate
Significant
Post-hoc analysis confirms that the ANN model significantly outperforms traditional models.
Confidence Interval Analysis
A 95% confidence interval was calculated for the mean prediction error.
Table 4.8: Confidence Interval for Mean Prediction Error
Statistic
Value
Mean Error
0.36
Lower Bound
-0.18
Upper Bound
0.90
The confidence interval includes zero, indicating that prediction errors are small and not statistically significant.
Early Warning Risk Classification
Based on predicted performance, students were categorized into risk levels.
Table 4.9: Early Warning Risk Classification Results
Risk Category
Number of Students
Percentage (%)
High Risk
96
8%
Moderate Risk
312
26%
Low Risk
792
66%
Total
1200
100%
The model successfully identified students at high academic risk. This demonstrates the practical applicability
of the predictive system in supporting early academic intervention strategies.
Feature Importance Analysis
To better understand the contribution of individual input variables to the predictive performance of the Artificial
Neural Network (ANN) model, a feature importance analysis was conducted. The relative importance of each
feature was estimated based on its contribution to reducing prediction error during model training. This analysis
helps identify the most influential academic and behavioral factors affecting student performance.
Table 4.10: Relative Importance of Input Features
Feature
Importance Score
Rank
Continuous Assessment
0.34
1
Attendance Percentage
0.26
2
Page 359
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
LMS Engagement
0.18
3
Assignment Scores
0.14
4
Midterm Marks
0.08
5
The results indicate that continuous assessment scores are the most influential predictor of final academic
performance, followed by attendance percentage and LMS engagement. These findings highlight the importance
of ongoing academic evaluation and student participation throughout the semester. Behavioral engagement
indicators, particularly LMS interaction, also play a meaningful role in forecasting outcomes. The relatively
lower contribution of midterm marks suggests that continuous performance monitoring provides stronger
predictive power than single examination results.
DISCUSSION OF FINDINGS
The experimental results indicate that the proposed Artificial Neural Network (ANN) model performs effectively
in predicting student academic performance. The model achieved high predictive accuracy on both the training
and testing datasets, with low error values and an R² of 0.88 on the testing set.
This demonstrates that the ANN successfully captured complex relationships among academic, behavioral, and
engagement-related variables.
The comparison with traditional machine learning models, such as Decision Tree, K-Nearest Neighbors (KNN),
and Support Vector Machine (SVM), confirms the superior performance of the ANN. While Decision Trees
provided interpretable rules and SVM achieved moderate accuracy, the ANN consistently delivered lower
prediction errors and better generalization.
This suggests that nonlinear interactions among variables, such as attendance, assignment completion, and
continuous assessment scores, are significant predictors of final academic outcomes and are best modeled with
neural networks.
Statistical validation further supports these findings. The paired sample t-test indicates no significant difference
between actual and predicted marks, while ANOVA confirms the superiority of the ANN model over other
approaches. The confidence interval analysis demonstrates that prediction errors are minimal and reliable, further
reinforcing the model’s credibility.
Analysis of feature influence revealed that continuous assessment scores, attendance percentage, and LMS
engagement were the most critical factors affecting performance. This emphasizes the importance of monitoring
ongoing student engagement and assessment outcomes, rather than relying solely on final examination results.
The early warning classification successfully identified students at high or moderate academic risk, enabling
timely intervention strategies. Such predictive insights can guide educators in implementing targeted support
programs, improving academic outcomes, and reducing dropout rates.
Overall, the findings confirm that the ANN-based predictive framework is both technically robust and practically
applicable for early identification of at-risk students. It demonstrates the advantages of integrating academic,
behavioral, and engagement data into a single predictive system.
Summary
This chapter presented detailed experimental results and statistical validation of the proposed ANN-based
academic performance prediction model. The results confirm high predictive accuracy, strong generalization
capability, and statistically validated superiority over traditional machine learning models. The developed
framework demonstrates both technical robustness and practical applicability for early identification of
academically at-risk students.
Page 360
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
CONCLUSIONS, RECOMMENDATIONS, AND LIMITATIONS
Conclusions
This study developed and validated an Artificial Neural Network (ANN)-based predictive framework for
forecasting student academic performance at an early stage. The findings confirm that a deep learning approach
integrating academic and behavioral indicators can reliably predict final examination outcomes and effectively
identify students at academic risk before performance declines.
The results demonstrate that multidimensional data—including continuous assessment scores, attendance
patterns, assignment performance, and LMS engagement—provide strong predictive capability when modeled
collectively. The feature importance analysis revealed that ongoing assessment and participation indicators are
more influential than single examination measures, emphasizing the value of continuous academic monitoring.
From a practical perspective, the proposed framework functions as an early warning system that supports
educators in identifying at-risk students and implementing timely interventions. By transforming institutional
data into actionable insights, the model assists academic administrators in improving resource allocation, student
support strategies, and overall performance management.
Scientifically, this research contributes to the field of educational data mining by demonstrating the effectiveness
of ANN-based predictive modeling combined with statistical validation techniques. The study provides a
scalable and empirically validated framework that can support the integration of predictive analytics into modern
higher education systems.
High Predictive Accuracy of ANN:
The ANN model demonstrated superior predictive capability compared to traditional machine learning models
such as Decision Tree, K-Nearest Neighbors, and Support Vector Machine. With a testing R² of 0.88 and low
prediction errors (MAE = 3.78, RMSE = 4.89), the model reliably forecasts final academic performance.
Importance of Multidimensional Data:
Integrating multiple factors, including continuous assessment scores, attendance, assignment completion, and
LMS engagement, significantly improved prediction accuracy. This highlights the importance of considering
both academic and behavioral variables rather than relying solely on examination results.
Effectiveness of Early Warning System:
The predictive framework successfully classified students into risk categories (high, moderate, low). The system
can identify at-risk students before final examinations, enabling timely academic intervention strategies such as
counseling, remedial classes, and personalized support.
Statistical Validation Confirms Reliability:
Hypothesis testing (paired t-test), ANOVA, and confidence interval analysis confirmed that the ANN predictions
are statistically consistent with actual outcomes, reinforcing the robustness and reliability of the model.
Practical Applicability:
The framework can serve as a decision-support tool for educators and administrators, assisting in early detection
of weak students, resource allocation, and the design of targeted academic support strategies.
Overall, the research demonstrates that ANN-based predictive models can effectively transform raw educational
data into actionable insights for improving student academic outcomes.
Page 361
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
Recommendations
Based on the findings, the following recommendations are proposed for educational institutions and future
research:
Implementation of Early Intervention Programs:
Institutions should integrate ANN-based predictive systems into their academic monitoring processes to identify
at-risk students and provide timely support.
Continuous Data Monitoring:
Regular collection of attendance, assessment, and LMS engagement data will ensure the predictive model
remains accurate and relevant over time.
Incorporation of Additional Variables:
Future implementations may include socio-economic, psychological, or demographic factors to further enhance
prediction accuracy and personalized support.
Training for Educators:
Faculty and administrative staff should be trained in interpreting predictive outputs and applying interventions
effectively without over-relying on automated predictions.
Scalability and Multi-Institutional Deployment:
Expanding the model across multiple courses or institutions can improve generalizability and enable
benchmarking of academic performance trends.
Limitations
Despite the positive outcomes, several limitations were identified:
Data Limitations:
The study relied on institutional datasets, which may contain incomplete or inconsistent records. Missing
variables or irregularities could influence prediction accuracy.
Context-Specific Findings:
The predictive model was developed based on data from a specific institution. Generalization to other institutions
may require additional tuning or retraining.
Exclusion of Non-Academic Factors:
Psychological, social, and family-related factors were not included. These may have a significant impact on
student performance and could further improve model accuracy if incorporated.
Temporal Dynamics Not Fully Explored:
The current model captures static patterns within a semester. Time-series modeling or recurrent neural networks
may enhance the ability to capture evolving learning behaviors over time.
Dependence on Data Quality:
ANN models require high-quality, consistent input data. Data entry errors or inconsistent engagement logs may
affect prediction reliability.
Page 362
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
Future Work
Building on this research, the following areas are recommended for future studies:
• Integration of longitudinal data to model changes in student behavior over multiple semesters.
• Exploration of advanced deep learning architectures (e.g., LSTM, GRU) to capture temporal dependencies
in student performance.
• Development of personalized learning recommendation systems based on early predictions.
• Inclusion of psychosocial, demographic, and peer influence factors to enrich prediction accuracy and
interpretability.
• Deployment of the system across multiple institutions to assess scalability and generalizability.
Summary
This chapter summarized the key findings of the study, highlighted practical and theoretical implications, and
provided actionable recommendations for educational institutions. While the ANN-based predictive framework
demonstrated high accuracy and practical utility, limitations were identified that can guide future research to
further enhance predictive performance and intervention strategies.
REFERENCES
1. R. Romero and S. Ventura, “Educational data mining: A review of the state of the art,” IEEE Transactions
on Systems, Man, and Cybernetics, Part C (Applications and Reviews), vol. 40, no. 6, pp. 601–618, Nov.
2010.
2. S. Yadav and A. Pal, “Data mining: A prediction for performance improvement of students using
classification,” International Journal of Computer Science and Information Technologies, vol. 5, no. 4,
pp. 5285–5289, 2014.
3. P. Cortez and A. Silva, “Using data mining to predict secondary school student performance,” Proceedings
of 5th Future Business Technology Conference, 2008, pp. 5–12.
4. S. Bhardwaj and S. Pal, “Data mining: Predicting academic performance of students using classification
techniques,” International Journal of Advanced Computer Science and Applications, vol. 3, no. 6, pp. 45–
49, 2012.
5. T. Kotsiantis, I. Pierrakeas, and P. Pintelas, “Predicting students’ performance in distance learning using
machine learning techniques,” Applied Artificial Intelligence, vol. 18, no. 5, pp. 411–426, 2004.
6. H. J. Lee, W. L. Chu, and K. H. Tan, “An ANN-based approach for predicting student performance,”
International Journal of Engineering Education, vol. 30, no. 2, pp. 293–303, 2014.
7. S. Al-Barrak and M. Al-Razgan, “Student performance prediction using machine learning techniques: A
systematic review,” International Journal of Advanced Computer Science and Applications, vol. 9, no. 6,
pp. 42–49, 2018.
8. S. M. Shafait, M. Shafique, and A. R. Malik, “Application of ANN for student performance prediction,”
Procedia Computer Science, vol. 57, pp. 791–797, 2015.
9. R. S. Baker and K. Yacef, “The state of educational data mining in 2009: A review and future visions,”
Journal of Educational Data Mining, vol. 1, no. 1, pp. 3–17, 2009.
10. S. Kotsiantis, “Use of machine learning techniques for educational proposes: A decision support system
for predicting students’ performance,” Artificial Intelligence Review, vol. 37, pp. 1–22, 2012.
11. P. Abdi and R. Sharma, “Comparative study of machine learning techniques for academic performance
prediction,” International Journal of Computer Applications, vol. 164, no. 3, pp. 1–6, 2017.
12. S. Pandey, A. Sharma, and P. Kumar, “Predicting student performance using ANN and SVM,”
International Journal of Computer Applications, vol. 179, no. 21, pp. 1–5, 2018.
13. R. K. Reddy and T. K. Varma, “A predictive model for student performance using artificial neural
networks,” Procedia Computer Science, vol. 132, pp. 834–841, 2018.
Page 363
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue II, February 2026
14. A. Jain, P. Kumar, and S. Gupta, “Educational data mining: A case study on student performance
prediction,” International Journal of Engineering and Technology, vol. 7, no. 3, pp. 216–221, 2018.
15. A. S. Safavian and D. Landgrebe, “A survey of decision tree classifier methodology,” IEEE Transactions
on Systems, Man, and Cybernetics, vol. 21, no. 3, pp. 660–674, 1991.
16. R. Hastie, T. Tibshirani, and J. Friedman, The Elements of Statistical Learning, 2nd ed. New York:
Springer, 2009.
17. D. Zhang, Q. Chen, and J. Chen, “A comparative study on machine learning algorithms for predicting
student performance,” Applied Soft Computing, vol. 89, p. 106100, 2020.
18. M. Al-Barrak, “Early warning systems in education: Machine learning approaches for at-risk students,”
International Journal of Educational Technology in Higher Education, vol. 17, no. 34, pp. 1– 22, 2020.
19. S. Z. Li and H. J. Xu, “Feature selection for predicting student academic performance: An ANN approach,”
Computers & Education, vol. 157, p. 103974, 2020.
20. R. K. Sharma and S. S. Patel, “Educational data mining and its applications: A review,” International
Journal of Advanced Research in Computer Science, vol. 11, no. 1, pp. 1–7, 2020.
