A Practical Model for Educators to Predict Student Performance in K-12 Education using Machine Learning

Full text

A Practical Model for Educators to Predict Student
Performance in K-12 Education using Machine Learning
Julie L. Harvey, Sathish A.P. Kumar
Coastal Carolina University
Conway, USA
{jlyantach, skumar} @coastal.edu
Abstract – Predicting classifiers can be used to analyze data
in K-12 education. Creating a classification model to
accurately identify factors affecting student performance
can be challenging. Much research has been conducted to
predict student performance in higher education, but there
is limited research in using data science to predict student
performance in K-12 education. Predictive models are
developed and examined in this review to analyze a K-12
education dataset. Three classifiers are used to develop
these predictive models, including linear regression,
decision tree, and Naive Bayes techniques. The Naive
Bayes techniques showed the highest accuracy when
predicting SAT Math scores for high school students. The
results from this review of current research and the models
presented in this paper can be used by stakeholders of K-12
education to make predictions of student performance and
be able to implement intervention strategies for students in
a timely manner.
Keywords: academic performance, data driven instruction,
data mining, education, intervention strategies, machine
learning, Massachusetts, predictive models
I. INTRODUCTION
Academic performance influences the performance of
students in their future careers. Students need successful
performance in education for the demands they will face in
their future careers. The dataset will be analyzed and
predictive models will be developed, correlating different
variables to student performance. The intended audience is
for those interested in the educational system. The method
for analyzing student performance indicators can be used by
data scientists, educators, administrators, parents, students,
and other stakeholders. Educators using models to identify
factors able to predict student performance in the K-12
classroom is not widely prevalent in the field of education.
The results of this study can be used in future studies to
evaluate K-12 student data for performance indicators and/or
predicting models.
Datamining and machine learning has been applied for
variety of fields [43-52]. Much research has be conducted
using data mining in the field of education, but the research
has largely been focused on higher education. There is little
research examining data mining for K-12 education. In this
paper, machine learning will be used to determine an
effective model for identifying factors affecting K-12 student
performance on SAT math scores. The model will also be
easy to use for educators and stakeholders. Using data
mining in education is time consuming and varies from state
to state and even from classroom to classroom. The goal of
all stakeholders in education is to help students become
successful now and in the future. Identifying features which
influence student performance, and educators using
identified information to direct instruction, will help to direct
educators, parents, administration and students to focus
attention on overcoming features which may hinder or
identifying features which improve student performance.
Interventions for improving SAT (standardized scholastic
aptitude test used for admitting undergraduate students in US
higher educational institutions) math scores can be put in
place earlier if data is examined at an earlier age for students.
This paper will demonstrate analysis of models used to
identify features leading to student performance, based on the
dataset to support the conclusion of the study.
The objective of this paper is to provide a method for
identifying the most indicative features of a successful
performance of students in grades K-12. This study will
provide a prediction model to be used on a variety of data sets
and how the data can be used to formulate solutions for
improving performance in K-12 education. Three classifiers;
linear regression technique, decision tree technique, and
Naive Bayes technique, will be evaluated to determine which
has the best prediction accuracy for the current dataset. These
classifiers were chosen for this study because of their
common use with education datasets. Choosing the best
prediction model for this problem can be complex because of
the amount of options there are for creating models and the
variables used in the dataset. There are also challenges in
choosing the model(s) best for identifying features of a
dataset that are most indicative of student success. Choosing
the most appropriate classifier will provide the most useful
prediction of factors affecting student performance. This
study is based on Massachusetts’s Public Schools dataset.
II. LITERATURE REVIEW
Most research of machine learning in the field of
education centers around the evaluation and analysis of data
from higher education. There are limited studies addressing
the use of prediction modeling in K-12 education. Recent
literature examines the use of different classification
techniques in predicting student performance. Kaur, et al [6]

examined data mining in secondary education. Slow learners
were identified among students using their classification
model. Five classifiers were used to predict the accuracy and
find the algorithm which will best for identifying students at-
risk in this data set: Multilayer Perception, Naive Bayes,
Sequential Minimal Optimization, J48 decision tree
algorithm, and Reduced Error Pruning Decision Tree all
using WEKA. Multilayer Perception showed the best
accuracy of 75% for the dataset. Primary school performance
predictions were made by Singh, et. al [42] of student
academic performance using previous exam results.
Researchers used decision tree, Naive Bayes, and Zero R
classification algorithms. The decision tree model showed a
slightly higher accuracy than the Naive Bayes model and both
the decision tree and Naive Bayes models had higher
accuracy than the Zero R model.
Meier, et. al [1] proposes a model to predict the final
grades of students in the class. Optimal prediction is learned
by the algorithm online, as well as the optimal time to issue
the prediction. These are based on students’ past
performances. The method was shown to be effective in
producing timely predictions to enable timely intervention by
the professor. Meier, et. al [21] used an online algorithm to
predict grades for each student in a particular class. The
algorithm they used increases in accuracy over the course of
the semester when using only the class data. The algorithm
also shows greater accuracy than linear regression and
decision tree when comparing models using the same data
set.
Rovira, et. al [3] uses 5 classifiers: Logistic Regression,
Gaussian Naive Bayes, Support Vector Machine, Random
Forest, and Adaptive Boosting. This article suggests a data-
driven system for extracting information from student
academic data that may be hidden, but is relevant to tutors of
the students. The model makes prediction about dropout
intention and student grades. Recommendation for courses
to take in the future is personalized for each student. A
variety of visualizations are presented of result interpretation.
Ramesh, et. al [16] examined 5 different classifiers for
predicting student performance: Naive Bayes, Multilayer
Perception, Sequential Minimal Optimization, J48 decision
tree, and Reduced Error Pruning Decision Tree. Multilayer
Perception showed the best accuracy of all of the models with
a 72.38% accuracy using the particular data set. Kavakchieva
[18] used WEKA classifiers to predict student performance
at the university level. J48 decision tree algorithm, Naive
Bayes, Bayes Net, k-nearest neighbor algorithm, and two rule
learners were used in this study. The J48 decision tree
classifier produced the best accuracy for this dataset, 66.59%.
Kotsiantis [27] compared different regression techniques to
forecast student grades: model trees, M5 rules algorithm,
artificial neural networks, locally weighted linear regression,
and sequential minimal optimization algorithm. The most
accurate algorithm was the M5 rules algorithm for this study,
and is also more comprehensible than the other models.
Gray, et. al [40] used academic and behavioral attributes to
predict student performance by evaluating and comparing 6
different classifiers. The classifiers used in this study were
Naive Bayes, unpruned decision tree, logistic regression,
support vector machine, neural network, and k-nearest
neighbor. SVM had an accuracy of 73.3%, which was the
highest accuracy of all of the models for this dataset.
Iqbal, et. al [2] used three different classification models
to predict student grades: Collaborative Filtering, Matrix
Factorization, and Restricted Boltzmann Machines
techniques. The intention was to improve the performance of
students and identify students who may need additional
support. The Restricted Boltzmann Machines technique was
found to be the best in predicting student performance in this
course. Tekin [19] used neural network, SVM, and ELM
algorithms to predict student grade point averages at
graduation. The most accurate results for this dataset came
from the SVM technique with a rate of 97.98%. The other
two models only had slightly less accuracy. Mishra, et. al
[39] used J48 decision tree and Random Tree to a data set of
college student to predict performance. Random Tree model
had better accuracy (94.418%) for the dataset used in this
study. Researchers used academic performance from
students’ first year of study to predict third semester
performance. Results from students’ second semester were
key to influencing the performance in third semester. They
also examined emotional attributes and found leadership and
drive to be significantly influential in student academic
performance.
Cheewaprakobkit [9] evaluated decision trees and neural
networks to classify student academic achievement. The
study also analyzes the factors affecting student academic
achievement. Attributes are analyzed using WEKA with
attribute importance being analyzed. Prediction accuracy
was evaluated using a 10-fold cross validation. The decision
tree resulted in higher accuracy than the neural network
classifier. Hamsa, et. al [11] used a C4.5 decision tree
algorithm and Fuzzy Genetic Algorithm to predict student
academic performance. Researchers analyzed students at risk
for failure. The decision tree algorithm projected more
students at risk because students on the barrier were
considered at risk. More students would require intervention,
whereas with the Fuzzy Genetic Algorithm students on the
cusp of at risk are considered safe and given mental
satisfaction. Livieris, et. al [29] used artificial neural
networks to predict student performance to incorporate a
broader range of student behavior into the dataset. Student
performance is a single course was evaluated. Four different
algorithms were used and compared for accuracy: Broyden-
Fletcher-Goldfarb-Shanno, Levenberg-Marquardt, Resilient
Backpropagation, and modified spectral Perry. The modified
spectral Perry showed the highest accuracy and was then
compared to C4.5 decision tree, Naive Bayes, Ripper, and
SMO algorithms where it still had the highest accuracy.
Ahmed and Elaraby [23] used ID3 decision tree to predict
student performance. The study examined individual
students in a college level mathematics class. Ogunde and
Ajibade [32] used ID3 decision tree to predict graduation
grades of university students. The model produced a

79.556% accuracy for the dataset used in the study.
Multilayer Perception ANN model was used by Oladokun, et.
al [26] to predict student performance in an engineering
course. They examine factors that may influence student
performance and were used in the model. The accuracy of
their model was over 74% and could be used to select
students seeking admission into a university. Christian and
Ayub [35] used Naive Bayes Tree classification technique to
predict student performance of university graduates. Many
different attributes were tested to determine whether they
were influential to student performance predictions. Gender
significantly influenced whether the student would finish
their studies. Credit had a significant influence on graduate
GPA. Test scores significantly affected the graduate’s length
of study. GPA significantly affected whether students
finished their studies.
Amrieh, et al [14] used artificial neural network, Naive
Bayes, and decision tree to classify data, but then used
Bagging, Boosting and Random Forest ensemble methods to
predict student performance. Teaching methods and learning
processes are said to be improved with the use of data mining
and being able to extract hidden knowledge from the
educational data. Enhancement of these processes leads to
improvements in student performance and overall
educational success. The authors propose a model to
examine student’s behavioral features. The model was
evaluated by classifiers and a strong relationship was shown
between the student’s behaviors and their achievement,
academically. The model is shown to be reliable with the
accuracy of the model being 22.1% with behavior features
and an accuracy of more than 80% when testing the model
with new students. Xu, et. al [4, 5] used an ensemble learning
technique to examine predictions of student performance in
college programs and the completion of a college degree.
The predictions are based on the students’ changing
performances. Previous academic records are examined by
Xu, et. al [4, 5] to accurately predict future performance of
students. They also allow for effectively providing students
with interventions for timely graduation. These studies
provide a novel algorithm for progressively predicting
student performance. Education-specific domain knowledge
is utilized and learning techniques are adapted. Livieris, et.
al [7, 29] present a new tool for prediction students’
performance on the final exams. The new decision support
tool is presented as user-friendly, with a simple interface used
in any platform and on any operating system. The suggested
decision support tool is a hybrid of a number of machine
learning methods aimed to achieve better performance than
any one method alone. The following algorithms were
combined for the ensemble method: Naive Bayes, ANN Back
Propagation algorithm, rule-learning technique RIPPER
algorithm, instance-based learner 3NN algorithm with
Euclidean distance, C4.5 decision tree algorithm, and the
Sequential Minimal Optimization SVM algorithm. Pandey
and Taruna [30] suggested a multilevel classification model
to first evaluate and then compared J48 decision tree, Lazy
Learner, Multilayer Perception neural network and Naive
Bayes. The model then uses a filtered dataset to enhance the
overall accuracy and the accuracy of each individual method.
J48 decision tree was found to be the best individual
predictive model, and the decision tree multilevel
classification model produced a 99.79% accuracy for the
dataset in this study.
III. METHODOLOGY
There are many different classifiers used to evaluate the
accuracy of a model using the dataset and other similar
datasets. Identifying an accurate predictive model is needed
for K-12 education datasets. Conducting data analysis by
education stakeholders is time consuming, costly process,
and does not always have a high level of accuracy. These
factors need to be improved to make educational data more
meaningful for students, teachers, and other stakeholders.
Accuracy of a prediction model may be worth a higher cost
if it provides educators with information that can be used to
help improve student performance. Analysis of the data can
be problematic if data is missing, data fields are empty, or if
there are errors in the model. The primary objective of this
report is to evaluate the most accurate classification
technique for predicting student performance. If the models
all display high levels of accuracy, cost and time will be
examined to determine the most effective model for
predicting student performance.
Figure 1. Methodology for the work
The data used for this report is the Massachusetts Public
Schools Data from the Massachusetts Department of
Elementary and Secondary Education website. Every year
the schools and districts in the state of Massachusetts are
issued a Report Card regarding the demographics, assessment
and accountability, educator information, and finances of the
schools in the state. Assessments are reported on student
performance on state standardized tests. The data was
collected from the Massachusetts State Department of
Elementary and Secondary Education and the Census
Bureau. Raw data files are provided on the state department
website. All of the data, including demographics, teacher
salaries, school finances, and numeric grades, can be
extracted from the website for data analysis. Information for
this study was collected from the state department website.
Student demographics, salaries of educators, and assessment
data were used in data analysis. The dataset consisted of
1861 observations and 303 variables. Each school in the state
is represented by 1 row. The dataset was cleaned to include
only high school data of 403 observations and the variables
were limited to 27 variables.

The data was cleaned, eliminating unnecessary data
variables and variables not being evaluated in this study. We
did eliminated some data about the teachers in this dataset
and variable involving the finances of the schools. We also
focused on specific test score data and eliminated the rest of
the test scores data. We selected variables specific to
individual students to examine their correlation with the
variable of interest. The variables are presented in the table
below.
Variable Abbreviation Variable Description
% African American Percentage of African American
students
Average Salary Average salary of the teachers
Avg_dist_exp Average expenditure per student
by district
Avg_exp Average expenditure per student
by school
% Non-Grad Completers Percentage of student that did not
graduate
% Males Percentage of males
% Dropped Out Percentage of student who
dropped out of school
% GED Percentage of students who
completed a GED
% MA Community College Percentage of students attending a
MA community college
Econ_disadvan Number of student with an
economic disadvantage
% Hispanic Percentage of Hispanic students
% Attending College Percentage of students attending
college
% Private Four-Year Percentage of students at a private
four-year college
% Public Four-Year Percentage of students at a public
four-year college
% Graduated Percentage of students who
graduated from the school
% White Percentage of White students
% AP_Score_3-5 Percentage of students with AP
score of 3-5
Average SAT_Reading Average SAT Reading score of
students
Average SAT_Math Average SAT Math score of
students
% Females Percentage of female students
Average Class Size Average size of class in school
% Asian Percentage of Asian students
TOTAL_Enrollment Total number of student in
attendance of the school
SAT_Tests_Taken Number of students who took the
SAT
First Language Not English Students whose first language is
not English
High Needs Student requiring additional
education accomodations
Economically Disadvantaged Number of students with an
economic disadvantage
Table 1. Variables Used in Data Analysis
Missing test scores were analyzed using ggplot. Missing
data was eliminated from the data set. The data was
subdivided into training and testing subsets. The training set
consisted of 70% of the original data set and the testing set
contained 30%. Data was displayed in scatter plots for the
linear regression model, an rpart tree for the decision tree
model, and a table was created using the Naive Bayes model.
All three models were used to make predictions about student
performance and the models were analyzed for accuracy.
IV. RESULTS
Schools having a higher percentage of economically
disadvantaged students tend to have missing SAT and AP
scores. Figure 2 shows the percentage of economically
disadvantaged students of all schools and Figure 3 shows the
percentage of economically disadvantaged students of
schools with missing scores.
Figure 2. Percentage of Economically disadvantaged students of
all schools
Figure 3. Percentage of Economically disadvantaged students of
schools with missing scores
K nearest neighbor was used to impute missing values.
The imputed and normalized data is shown in Figure 4.
Figure 4. Imputed and Normalized Data

Linear regression was the first classification technique
used to examine this data set. SAT Math scores were chosen
as the variable of interest for all classification techniques. AP
scores, SAT Reading scores, and SAT Math scores are all
highly correlated, with any of these variables able to be used
for evaluation. Correlation plot and corrgram were created to
determine correlation between the 27 variables in Figures 5
and 6.
Figure 5. Correlation plot
Figure 6. Corrgram
Expenditures per pupil is highly correlated with
teachers’ salary, but not with class size. The average SAT
math scores display a slightly negative correlation with
expenditure. Schools with a larger number of high needs
students correlates with higher expenditures. Socioeconomic
factors are shown to correlate with students’ SAT scores. The
following figures displays the relationships between average
SAT math scores and the following variables; the percentage
of economically disadvantaged, teacher average salary,
average class size, average expenditure per pupil, and total
student enrollment. The percentage of economically
disadvantaged students shows a strong negative correlation
with average SAT math scores as shown in Figure 7.
Figure 7. Percentage of Economically Disadvantaged Students
Versus Average SAT Math Scores
The scatter plots in Figures 8 to 11 show a stronger
relationship between average SAT math scores and
expenditures and total enrollment than between the average
SAT math scores and average teacher salary and average
class size.
Figure 8. Scatter plot for average expenditure per pupil vs. average
SAT math scores.
Figure 9. Scatter plot for total enrollment vs. average SAT math
scores.

Figure 10. Scatter plot for average salary vs. average SAT math
scores.
Figure 11. Scatter plot for average class size vs. average SAT
math scores.
Linear regression analysis was used to examine
expenditure per pupil, total enrollment, average salary and
average class size while controlling for the percentage of
economically disadvantaged students. Average expenditure
is better at explaining variation in average SAT math than
average teacher salary when controlling for economic
disadvantage. Both variables show a negative relationship
with one standard deviation increase in average expenditure
per pupil associated with a 0.083 standard deviation decrease
in average SAT math scores. Schools with more funding tend
to have a lower socioeconomic level, which correlates with
lower SAT scores. This could explain why the higher
expenditure per pupil is negatively correlated with average
SAT math scores. Average class size is better at explaining
variation in average SAT math scores than total student
enrollment. The scatter plot showed a stronger relationship
between average SAT math scores and total enrollment than
the linear regression model. This may be because schools
with a higher total student enrollment tend to have a lower
percentage of economically disadvantaged students. The
final linear regression included the percentage of students
who are economically disadvantaged, average expenditure
per pupil, and average class size. The average SAT math
scores decreases by approximately 0.78 standard deviation on
average with one standard deviation increase in the
percentage of economically disadvantaged students, with
class size and expenditure kept constant. Accuracy of the
model was tested. The model summary returned a perfect fit,
with the summary possibly being unreliable. The plots of the
residuals is shown in Figures 12 and 13.
A data set was created of the actual and predicted results
to check the performance of the model. As shown in Figure
14, the current performance of the model is 1. The model is
shown to explain 100% of variance in the test data.
The decision tree classification technique was another
model developed to analyze the Massachusetts data set. 70%
of the observations were randomly chosen for the training
subset and 30% for the testing subset. As seen in Figure 15,
a decision tree was created using the rpart and rattle libraries
in R Studio.
Figure 12. Histogram plot for residuals
Figure 13. Scatter plot for residuals
Figure 14. Model performance evaluation

Figure 15. Decision Tree classifier implementation
The model was used to predict performance on the SAT
math test and evaluate the prediction. As seen in Figure 16,
a confusion matrix was created and demonstrated the
accuracy of the model to be around 59.8%
Figure 16. Confusion Matrix to test the model performance
The Naive Bayes classification technique was also used
to evaluate this education data set. The data set was split into
70% training subset and 30% of the data into a testing subset.
The classification model was built using the training subset
and the model was used to predict performance on the SAT
math test. Figure 17 displays the prediction results of the
model.
Figure 17. Prediction results for the Naïve Bayes Classifier.
The accuracy of the naïve Bayes classifier model was
71.0%. This was the most effective classification technique
used in this study for the prediction of SAT math test scores.
Selection of classification techniques to apply to a dataset
requires analysis of the dataset. There are numerous
classifiers used to analyze educational data and the dataset
determines which classifier is best for analyzing the data.
The intention of the analysis also needs to be considered
when choosing the best classifier for a dataset. Cleaning the
large amount of data available to educators is a time
consuming and costly process. The Naive Bayes model built
in this study can be easily and quickly used for an individual
classroom, or on a large scale such as the state department
dataset used here.
V. CONCLUSION
The Naive Bayes classification model can be used to
make predictions of other performance measures, using the
same data set. Educators can use the model to choose
variables they want to evaluate correlation between and
determine how the variables are used to predict student
performance on assessments. This prediction model can then
be used to implement early intervention for students, which
can help to improve student future success. Timely feedback
is important for educators to be able to provide early
intervention strategies. This model can be implemented
quickly and easily once the data set is loaded.
Future research can be conducted to use more
performance assessment measures to determine future
performance of students. The linear regression model can be
improved upon to make better predictions of student
performance. It would be interesting to compare the
performance of the classification models on other datasets, or
even improving the model to increase prediction accuracy of
the current model. Future research could also include
evaluating behavioral characteristics of students, and other
demographics as they relate to student performance on
assessment measures. Additional analysis could be
conducted using different classifiers on the same dataset,
including multilayer perception and artificial neural network.
REFERENCES
[1] Meier, Y., Xu, J., Atan, O., & Van Der Schaar, M. (2015, November).
Personalized grade prediction: A data mining approach. In Data Mining
(ICDM), 2015 IEEE International Conference on (pp. 907-912). IEEE.
[2] Iqbal, Z., Qadir, J., Mian, A. N., & Kamiran, F. (2017). Machine Learning
Based Student Grade Prediction: A Case Study. arXiv preprint
arXiv:1708.08744.
[3] Rovira, S., Puertas, E., & Igual, L. (2017). Data-driven system to predict
academic grades and dropout. PloS one, 12(2), e0171207.
[4] Xu, J., Moon, K. H., & Van Der Schaar, M. (2017). A Machine Learning
Approach for Tracking and Predicting Student Performance in Degree
Programs. IEEE Journal on Selected Topics in Signal Processing, 11(5), 742-
753.
[5] Xu, J., Han, Y., Marcu, D., & Van Der Schaar, M. (2017, February).
Progressive Prediction of Student Performance in College Programs. In
AAAI (pp. 1604-1610).
[6] Kaur, P., Singh, M., & Josan, G. S. (2015). Classification and prediction
based data mining algorithms to predict slow learners in education sector.
Procedia Computer Science, 57, 500-508.
[7] Livieris, I. E., Mikropoulos, T. A., & Pintelas, P. (2016). A decision
support system for predicting students’ performance. Themes in Science and
Technology Education, 9(1), 43-57.
[8] Oviedo, B., Moral, S., & Puris, A. (2016). A hierarchical clustering
method: Applications to educational data. Intelligent Data Analysis, 20(4),
933-951.
[9] Cheewaprakobkit, P. (2015). Predicting Student Academic Achievement
by Using the Decision Tree and Neural Network Techniques. Catalyst, 12(2).
[10] Mohamad, S. K., & Tasir, Z. (2013). Educational data mining: A review.
Procedia-Social and Behavioral Sciences, 97, 320-324.

[11] Hamsa, H., Indiradevi, S., & Kizhakkethottam, J. J. (2016). Student
academic performance prediction model using decision tree and fuzzy
genetic algorithm. Procedia Technology, 25, 326-332.
[12] Shahiri, A. M., & Husain, W. (2015). A review on predicting student's
performance using data mining techniques. Procedia Computer Science, 72,
414-422.
[13] Kotsiantis, S., Pierrakeas, C., & Pintelas, P. (2004). Predicting students’
performance in distance learning using machine learning techniques.
Applied Artificial Intelligence, 18(5), 411-426.
[14] Amrieh, E. A., Hamtini, T., & Aljarah, I. (2016). Mining educational
data to predict Student’s academic performance using ensemble methods.
International Journal of Database Theory and Application, 9(8), 119-136.
[15] Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., &
Salakhutdinov, R. (2014). Dropout: A simple way to prevent neural networks
from overfitting. The Journal of Machine Learning Research, 15(1), 1929-
1958.
[16] Ramesh, V., Parkavi, P., & Ramar, K. (2013). Predicting student
performance: a statistical and data mining approach. International journal of
computer applications, 63(8).
[17] Cortez, P., & Silva, A. M. G. (2008). Using data mining to predict
secondary school student performance.
[18] Kabakchieva, D. (2013). Predicting student performance by using data
mining methods for classification. Cybernetics and information
technologies, 13(1), 61-72.
[19] Tekin, A. (2014). Early Prediction of Students' Grade Point Averages at
Graduation: A Data Mining Approach. Eurasian Journal of Educational
Research, 54, 207-226.
[20] Thai-Nghe, N., Horváth, T., & Schmidt-Thieme, L. (2011, July).
Factorization Models for Forecasting Student Performance. In EDM (pp. 11-
20).
[21] Meier, Y., Xu, J., Atan, O., & Van der Schaar, M. (2016). Predicting
grades. IEEE Transactions on Signal Processing, 64(4), 959-972.
[22] Rathee, A., & Mathur, R. P. (2013). Survey on decision tree
classification algorithms for the evaluation of student performance.
International Journal of Computers & Technology, 4(2a1), 244-247.
[23] Ahmed, A. B. E. D., & Elaraby, I. S. (2014). Data Mining: A prediction
for Student's Performance Using Classification Method. World Journal of
Computer Application and Technology, 2(2), 43-47.
[24] Baker, R. S., & Inventado, P. S. (2014). Educational data mining and
learning analytics. In Learning analytics (pp. 83-98). Springer New York.
[25] Chen, J. F., & Do, Q. H. (2014). Training neural networks to predict
student academic performance: A comparison of cuckoo search and
gravitational search algorithms. International Journal of Computational
Intelligence and Applications, 13(01), 1450005.
[26] Oladokun, V. O., Adebanjo, A. T., & Charles-Owaba, O. E. (2008).
Predicting students’ academic performance using artificial neural network:
A case study of an engineering course. The Pacific Journal of Science and
Technology, 9(1), 72-79.
[27] Kotsiantis, S. B. (2012). Use of machine learning techniques for
educational proposes: a decision support system for forecasting students’
grades. Artificial Intelligence Review, 37(4), 331-344.
[28] Chen, J. F., Hsieh, H. N., & Do, Q. H. (2014). Predicting student
academic performance: a comparison of two meta-heuristic algorithms
inspired by cuckoo birds for training neural networks. Algorithms, 7(4), 538-
553.
[29] Livieris, I. E., Drakopoulou, K., & Pintelas, P. (2012). Predicting
students' performance using artificial neural networks. In 8th PanHellenic
Conference with International Participation Information and
Communication Technologies in Education (pp. 321-328).
[30] Pandey, M., & Taruna, S. (2014). A Multi-level Classification Model
Pertaining to The Student's Academic Performance Prediction. International
Journal of Advances in Engineering & Technology, 7(4), 1329.
[31] Guarín, C. E. L., Guzmán, E. L., & González, F. A. (2015). A model to
predict low academic performance at a specific enrollment using data
mining. IEEE Revista Iberoamericana de tecnologias del Aprendizaje, 10(3),
119-125.
[32] Ogunde, A. O., & Ajibade, D. A. (2014). A data mining system for
predicting university students’ graduation grades using ID3 decision tree
algorithm. Journal of Computer Science and Information Technology, 2(1),
21-46.
[33] Mayilvaganan, M., & Kalpanadevi, D. (2014, December). Comparison
of classification techniques for predicting the performance of students
academic environment. In Communication and Network Technologies
(ICCNT), 2014 International Conference on (pp. 113-118). IEEE.
[34] Natek, S., & Zwilling, M. (2014). Student data mining solution–
knowledge management system related to higher education institutions.
Expert systems with applications, 41(14), 6400-6407.
[35] Christian, T. M., & Ayub, M. (2014, November). Exploration of
classification using NBTree for predicting students' performance. In Data
and Software Engineering (ICODSE), 2014 International Conference on (pp.
1-6). IEEE.
[36] Jishan, S. T., Rashu, R. I., Haque, N., & Rahman, R. M. (2015).
Improving accuracy of students’ final grade prediction model using optimal
equal width binning and synthetic minority over-sampling technique.
Decision Analytics, 2(1), 1.
[37] Tucker, C., Pursel, B. K., & Divinsky, A. (2014). Mining student-
generated textual data in MOOCs and quantifying their effects on student
performance and learning outcomes. Computers in Education Journal, 5(4),
84-95.
[38] Bogarín, A., Romero, C., Cerezo, R., & Sánchez-Santillán, M. (2014,
March). Clustering for improving educational process mining. In
Proceedings of the Fourth International Conference on Learning Analytics
And Knowledge (pp. 11-15). ACM.
[39] Mishra, T., Kumar, D., & Gupta, S. (2014, February). Mining students'
data for prediction performance. In Advanced Computing & Communication
Technologies (ACCT), 2014 Fourth International Conference on (pp. 255-
262). IEEE.
[40] Gray, G., McGuinness, C., & Owende, P. (2014, February). An
application of classification models to predict learner progression in tertiary
education. In Advance Computing Conference (IACC), 2014 IEEE
International (pp. 549-554). IEEE.
[41] Saa, A. A. (2016). Educational Data Mining & Students’ Performance
Prediction. International Journal of Advanced Computer Science and
Applications, 7(5), 212-220.
[42] Singh, M., Nagar, H., & Sant, A. (2016) Using Data Mining to Predict
Primary School Student Performance. IJARIIE-ISSN(O)-2395-4396, 2(1),
43-46.
[43] B, Dahal, Kumar, S.A.P. & Li, Z, "Topic modeling and sentiment
analysis of global climate change tweets", Soc. Netw. Anal. Min. (2019) 9:
24.
[44] SAP Kumar, R. Madhumathi, P.R. Chelliah, L. Tao, and S. Wang, "A
novel digital twin-centric approach for driver intention prediction and traffic
congestion avoidance.", Journal of Reliable Intelligent Environments 4, no.
4 (2018): 199-209.
[45] S.K. Alampalayam, and Essam F. Natsheh, "Multivariate Fuzzy
Analysis for Mobile Ad hoc Network Threat Detection.", International
Journal of Business Data Communications and Networking 4, no. 3 (2008):
1.
[46] B. Xu, and S.A. Kumar,"Big Data Analytics Framework for Improved
Decision Making.", In Proceedings on the International Conference on
Internet Computing (ICOMP), p. 1.
[47] S. Alampalayam, A. Kumar, J. H. Graham, and S. Sri nivasan, "Intruder
Identification and Response Framework for Mobile Ad hoc Networks."
In Computers and Their Applications, pp. 260-265. 2007.
[48] SAP, Kumar, and B. Xu, "A Machine Learning Based Approach to
Detect Malicious Fast Flux Networks", In 2018 IEEE Symposium Series on
Computational Intelligence (SSCI), pp. 1676-1683. IEEE, 2018.
[49] S. Kumar, S. Bao, V. Singh, and J. Hallstrom, "Flooding disaster
resilience information framework for smart and connected communities",
Journal of Reliable Intelligent Environments 5, no. 1 (2019).
[50] S. Singleton, S. A. P. Kumar and Z. Li, "Twitter Analytics-Based
Assessment: Are the United States Coastal Regions Prepared for Climate
Changeƒ,", 2018 IEEE International Symposium on Technology and Society
(ISTAS), Washington DC, DC, USA, 2018, pp. 150-155.
[51] P. Raj and SAP Kumar, "Big Data Analytics Processes and Platforms
Facilitating Smart Cities", Smart Cities: Foundations, Principles, and
Applications (2017): 23-52.
[52] D. Eastman, and SAP Kumar. "A Simulation Study to Detect Attacks
on Internet of Things", In 2017 IEEE 15th Intl Conf on Dependable,
Autonomic and Secure Computing, pp. 645-650. IEEE, 2017.