Authors :
Adetoba, B. T.; Awodele, O.; Alao, O. D.; Nwaocha, V. O.
Volume/Issue :
Volume 5 - 2020, Issue 3 - March
Google Scholar :
https://goo.gl/DF9R4u
Scribd :
https://bit.ly/2wFmjFB
Abstract :
Classification algorithms have been found to
produce a better result for monitoring e-examination in
terms of performance and accuracy in detecting
examination impersonation in an e-learning
environment. This paper presents the results of the
comparative analysis of six classification algorithms
(Logistic Regression, Multi-Layer Perceptron, Support
Vector Machine, Random Forest, Bayes Network and
Stochastic Gradient Descent) for face matching and
verification in e-examination. This were compared and
evaluated based on True Positive Rate (TPR), False
Positive Rate (FPR), Precision, Recall, F-measure,
Kappa Statistics (KS), Accuracy, Time to build model
and Mean Absolute Error, using Waikato Environment
for Knowledge Analysis (WEKA) to determine the best
fit classifiers for the design of the model. The developed
model was tested, using 250 facial images (dataset)
acquired from the entire National Diploma students of
Computer Science, Yaba College of Technology. The
best fit classifiers (LR-based on Logistic Loss Function
and SGD-Stochastic Gradient Descent) obtained from
the comparison was used for binary classification,
image optimization and monitoring. The results from
the comparison showed that LR and SGD had leading
performances with TPR, FPR, Precision, Recall, F-
Measure, KS and Accuracy values of 100%, Mean
Absolute Error value of zero. LR had shortest time of
0.01 second and SGD with 0.04 second, based on time
taken to build model. LLF of LR and SGD played
significance role in producing faster and optimal face
recognition results. The technique can be employed by
examiners and learning management specialists to
conduct malpractice free e-examination.
Keywords :
Classification Algorithms; E-Examination; Face Matching; Face Verification; E-Learning Environment.
Classification algorithms have been found to
produce a better result for monitoring e-examination in
terms of performance and accuracy in detecting
examination impersonation in an e-learning
environment. This paper presents the results of the
comparative analysis of six classification algorithms
(Logistic Regression, Multi-Layer Perceptron, Support
Vector Machine, Random Forest, Bayes Network and
Stochastic Gradient Descent) for face matching and
verification in e-examination. This were compared and
evaluated based on True Positive Rate (TPR), False
Positive Rate (FPR), Precision, Recall, F-measure,
Kappa Statistics (KS), Accuracy, Time to build model
and Mean Absolute Error, using Waikato Environment
for Knowledge Analysis (WEKA) to determine the best
fit classifiers for the design of the model. The developed
model was tested, using 250 facial images (dataset)
acquired from the entire National Diploma students of
Computer Science, Yaba College of Technology. The
best fit classifiers (LR-based on Logistic Loss Function
and SGD-Stochastic Gradient Descent) obtained from
the comparison was used for binary classification,
image optimization and monitoring. The results from
the comparison showed that LR and SGD had leading
performances with TPR, FPR, Precision, Recall, F-
Measure, KS and Accuracy values of 100%, Mean
Absolute Error value of zero. LR had shortest time of
0.01 second and SGD with 0.04 second, based on time
taken to build model. LLF of LR and SGD played
significance role in producing faster and optimal face
recognition results. The technique can be employed by
examiners and learning management specialists to
conduct malpractice free e-examination.
Keywords :
Classification Algorithms; E-Examination; Face Matching; Face Verification; E-Learning Environment.
