Journal of Intelligent Learning Systems and Applications, 2023, 15, 67-75

https://www.scirp.org/journal/jilsa
ISSN Online: 2150-8410
ISSN Print: 2150-8402

Skin Cancer Classification Using Transfer

Learning by VGG16 Architecture (Case Study
on Kaggle Dataset)

Adam M. Ibrahim1*, Mohammed Elbasheir2*, Somia Badawi3, Ashraf Mohammed4,

Amir F. Mohammed Alalmin5
1
Department of Computer Science, Sudan Technical University, Khartoum, Sudan
2
Faculty of Computer Science and Information Technology, Alzaiem Alazhri University, Khartoum, Sudan
3
Applied College, Najran University, Najran, Saudi Arabia
4
Faculty of Computer Science and Information Technology, Omdurman Islamic University, Omdurman, Sudan
5
Department of Emergency Medicine, King Abdullah Medical City, Makkah, Saudi Arabia

How to cite this paper: Ibrahim, A.M., Abstract

Elbasheir, M., Badawi, S., Mohammed, A.
and Alalmin, A.F.M. (2023) Skin Cancer Skin cancer is a serious and potentially life-threatening disease that affects
Classification Using Transfer Learning by millions of people worldwide. Early detection and accurate diagnosis are crit-
VGG16 Architecture (Case Study on Kag-
ical for successful treatment and improved patient outcomes. In recent years,
gle Dataset). Journal of Intelligent Learning
Systems and Applications, 15, 67-75. deep learning has emerged as a powerful tool for medical image analysis, in-
https://doi.org/10.4236/jilsa.2023.153005 cluding the diagnosis of skin cancer. The importance of using deep learning
in diagnosing skin cancer lies in its ability to analyze large amounts of data
Received: April 1, 2023
Accepted: August 5, 2023
quickly and accurately. This can help doctors make more informed decisions
Published: August 8, 2023 about patient care and improve overall outcomes. Additionally, deep learning
models can be trained to recognize subtle patterns and features that may not
Copyright © 2023 by author(s) and
be visible to the human eye, leading to earlier detection and more effective
Scientific Research Publishing Inc.
This work is licensed under the Creative treatment. The pre-trained Visual Geometry Group 16 (VGG16) architecture
Commons Attribution International has been used in this study to classification of skin cancer images, and the
License (CC BY 4.0). images have been converted into other color scales, there are named: 1) Hue
http://creativecommons.org/licenses/by/4.0/
Saturation Value (HSV), 2) YCbCr, 3) Grayscale for evaluation. Results show
Open Access
that the dataset created with RGB and YCbCr images in field condition was
promising with a classification accuracy of 84.242%. The dataset has also been
evaluated with other popular architectures and compared. The performance
of VGG16 with images of each color scale is analyzed. In addition, feature
parameters have been extracted from the different layers. The extracted layers
were felt with the VGG16 to evaluate the ability of the feature parameters in
classifying the disease.

DOI: 10.4236/jilsa.2023.153005 Aug. 8, 2023 67 Journal of Intelligent Learning Systems and Applications
A. M. Ibrahim et al.

Keywords
Skin Cancer, Classification, VGG16, Transfer Learning, Deep Learning

1. Introduction
One of the common [1] and the famous types of cancer disease is that it comes
out as a result of DNA damage which later on causes or leads to mortality. Da-
maging DNA sets out cells to rise up out of control while the phenomenon is
exacerbated increasingly at present. Researchers conducted studies on compute-
rized images as analysis and sorting of the inner factors that stand beyond the
cancer case. The output is automated introducing that recognized skin cancer.
Reality of this automation proved the accuracy and precision of discovering the
sick spots-king in at most phases. Since cancer [2] is such a deadly illness, and is
regarded as a life threat. Additionally, cancer may afflict skin of human beings.
Even skin cancer is considered as a fastest growing disease that can inevitably
lead to death.
The most common [3] types of skin cancer are: keratosis, carcinoma, squam-
ous and melanoma. Health organization (WHO) is reported to claim that one in
every three cases of examined cancer is body skin one. In general, skin cancer is
observed to spread much amidst USA, Canada, Australia people that is accord-
ing to reports from Canadian Cancer Society’s Committee, Cancer Council in
Australia and WHO during the years 2014-2019 [4]. Cancer is an extremist life
threat to human life. It may sometimes cause certain death to the human. Dif-
ferent types of cancer may exist in the human body and skin cancer is one of the
fastest-growing cancers that can cause death. It is provoked by some factors like
smoking, alcohol usage, allergies, infections, viruses, physical activity, environ-
mental change, exposure to ultraviolet (UV) light, and so on. The DNA inside
the skin cells can be annihilated by the radiation of UV rays from the sun. In ad-
dition, unusual swellings of the human body are also a cause of skin cancer.
Deep learning techniques [5] for skin cancer detection: A great role is played
by deep neural networks in detecting skin cancer. In these nets there are groups
that mimic the human brain in shape. These nerves (The interconnected nets)
work with each in a harmoniously cooperative mechanism to troubleshoot stated
problems. Neural networks are trained to sortify, classify and analyze images of
skin cancer x-rays. Datasets are prepared out of international skin imaging col-
laboration (ISIC). The recognition software simulates different types of learning
such as ANN, CNN, etc. Of them, [6] the nonlinear and statistical method is an
intelligent network built like the biological brain; is so-called ANN. The system
is based on tri-layer neurons: the input, intermediate; the hidden layers then the
data is sent to the third layer of output neurons.
This study [7] [8] features a deep convolution neural network (DCNN) model

DOI: 10.4236/jilsa.2023.153005 68 Journal of Intelligent Learning Systems and Applications

 A. M. Ibrahim et al.

built on deep learning theory. The researchers use kernel to mitigate distraction-
al noise and artifacts. Classification depends on sorting and extracting images.
Augmentation of data enhances the precision of classification rate. Comparing
our model [9] showed here is the very mechanism to estimate (VGG16) being
similarized with some other learning models and views as CNN, AlexNet, Res-
Net, DenseNet. However, in this article we use Pre-Trained Visual Geometry
Group 16 (VGG16) architecture model.
Transfer Learning [10] is a machine learning technique where knowledge
gained from solving one problem can be used to solve another related problem.
However, it involves taking the weights and parameters learned on an existing
model, such as a convolutional neural network (CNN), and using them in order
to create new models that are better suited for different tasks. This approach al-
lows us to use fewer resources while still achieving good results with our models
[11].
VGG16 is a convolutional neural network (CNN) architecture [12] developed
by the Visual Geometry Group at Oxford University. It consists of 16 layers, in-
cluding 13 convolutional and 3 fully connected layers. The VGG16 model has
been used in many applications such as image classification, object detection,
segmentation etc., Transfer Learning with this architecture involves taking the
weights and parameters learned on an existing model using VGG16 and apply-
ing them to create new models that are better suited for different tasks.
Deep learning [13] has revolutionized the entire landscape of machine learn-
ing during recent decades. It is considered the most sophisticated machine
learning subfield concerned with artificial neural network algorithms. These al-
gorithms are inspired by the function and structure of the human brain. Deep
learning techniques are implemented in a broad range of areas such as speech
recognition, pattern recognition [14], and bioinformatics [15]. As compared
with other classical approaches of machine learning, deep learning systems have
achieved impressive results in these applications. Various deep learning ap-
proaches have been used for computer-based skin cancer detection in recent
years. In this paper, we thoroughly discuss and analyze skin cancer detection
techniques based on deep learning.

2. Proposed Skin-Cancer-Malignant Model Transfer

Learning by VGG16 Architecture
This paper contributes to suggest Skin-Cancer-Malignant software model sorts
and classifies skin cancer. This model is proved as accurate mechanism. Pro-
cessing the case by this model is a little bit sort-timed, besides adding a new layer
in VGG16 architecture than other soft models. In this work [16], a transfer
learning model known as ResNet50 was proposed, but without feature selection
or preprocessing steps. The results may be improved with a better classification
rate of skin injuries images. In [17], the author proposed a CNN-based model
for melanoma skin cancer detection using pre- and post-processing of images to
optimize segmentation. The model produced lesion areas by combining local

DOI: 10.4236/jilsa.2023.153005 69 Journal of Intelligent Learning Systems and Applications

A. M. Ibrahim et al.

and global contextual information. He got a good score for prediction and clas-
sification. However, execution time is not mentioned which may add value to
the results.
In [18] the author used CNN to detect melanoma from pigmented melanocyte
lesions from endoscopic images. However, non-melanoma and non-pigmented
melanomas were difficult to examine. It also has lower detection accuracy.
Besides, Skin-Cancer-Malignant model gives the most classification precision
and good execution time.
The researchers used Transfer Learning with the VGG16 architecture involves
taking the weights and parameters learned on an existing model using VGG16
and applying them to create new models that are better suited for skin cancer
classification. This approach allows us to use fewer resources while still achiev-
ing good results with our models. Figure 1 illustrates the process of skin cancer
detection using the VGG16 model.
For our classification, The VGG16 model was trained with the new dataset
firstly. Then, taken a pre-trained network such as VGG16 which has already
been trained on millions of images from ImageNet dataset, then fine tune it by
adding additional layers and changing some of its hyper parameters in order to
make it more suitable for our task at hand. The architecture of our model is
shown in Figure 2.

Figure 1. The process of skin cancer detection, VGG16: Pre-trained visual geometry group 16
architecture model.

DOI: 10.4236/jilsa.2023.153005 70 Journal of Intelligent Learning Systems and Applications

 A. M. Ibrahim et al.

Figure 2. The architecture of VGG16 model.

Our data set used in our model was Skin_Cancer_Malignant Vs Benign from
Kaggle dataset, this dataset contains training collection which contain of 1440
benign images 1197 malignant, these datasets used in VGG16 model as pre-train
dataset firstly. Then testing collection of 360 benign images and 300 of malig-
nant images which used as a test dataset after the VGG16 model run. Figure 3
explains the randomly selected samples of skin images from the dataset used be-
fore, while and after pre-processing and enhancement, starting with Clear im-
ages (a) Clear raw skin image with expected cancer, (b) Less Contrast, (c) Noise
and (d) Hair and Noise.

3. Experiment and Result Analysis

Transfer learning is a technique in machine learning where a pre-trained model
is used as a starting point for a new task. In the case of skin cancer classification,
the VGG16 model is often used as the pre-trained model.
The VGG16 architecture consists of 13 convolutional layers and 3 fully con-
nected layers. The convolutional layers are responsible for extracting features
from the input image, while the fully connected layers are responsible for mak-
ing predictions based on those features. To use VGG16 for skin cancer classifica-
tion, the last layer of the model (which is typically a softmax layer) is replaced
with a new output layer that has two nodes (one for malignant and one for be-
nign). The weights of all other layers in the model are frozen so that they do not
change during training.
The VGG16 model was tested on the CIFAR-10 dataset and achieved an ac-
curacy of 92.5%. This result is very impressive, considering that the dataset con-
tains 10 classes with 6000 images per class. The model was able to accurately
classify images with high accuracy, demonstrating its effectiveness in image rec-
ognition tasks. Furthermore, the model was able to achieve good result with a
relatively small number of parameters. This indicates that the VGG16 model is
an efficient and powerful tool for image recognition tasks.
The equation for this model is:
Y = β 0 + β1 X 1 + β 2 X 2 +  + β n X n

where Y is the dependent variable, X 1 , X 2 ,, X n are the independent variables,

and β 0 , β1 ,, β n are the coefficients.
When the model was trained, the parameters of the equations (e.g. weights
and biases) will change to better fit the data. This can be introduced by showing

DOI: 10.4236/jilsa.2023.153005 71 Journal of Intelligent Learning Systems and Applications

A. M. Ibrahim et al.

Figure 3. Samples of skin images used in Skin_Cancer_Malignant model. (a) Clear, (b)
Less contrast, (c) Noise, (d) Hair & Noise.

how the parameters change as the model is trained, and how this affects the equ-
ations.
The process of determining the weights of the layers in a VGG16 model used
to classify our skin cancer images involves pre-train the model on a dataset of
labeled images (training data 80%). This is done by using a supervised learning
algorithm named backpropagation. During this process, the weights of the layers
are adjusted to minimize the error between the predicted and actual labels for
each image. The weights are adjusted based on how well they contribute to cor-
rectly classifying each image. This process is repeated until the model reaches
high ACC with last epoch with our train data, then researchers used the weights
to classify the teste data 20% of remaining images which achieved ACC of
84.242.
The input to the model is an image of a skin lesion, which is passed through
the convolutional layers to extract features. These features are then flattened and
passed through the fully connected layers to make a prediction about whether or
not the lesion is malignant or benign. During training, only the weights of the
new output layer are updated using backpropagation. This allows us to train a
new model for skin cancer classification using much less data than would be re-
quired if we were training from scratch.
When this model is estimated, the obtain is 84.242% being standardized on
556 datasets.
Overall, the model is proved dependable and robust. One of the main advan-
tages of using transfer learning with VGG16 for skin cancer classification is that
it allows us to achieve high accuracy with relatively little data. This is because the
network has already learned to recognize many features that are relevant to skin
cancer classification, such as texture, color, and shape. Another advantage of
using transfer learning with VGG16 is that it can help us avoid overfitting. Over-
fitting occurs when a model becomes too complex and starts to memorize the
training data instead of generalizing to new data. By starting with a pre-trained
model like VGG16, we can avoid overfitting by leveraging the knowledge that
the network has already learned.
Figure 4 shows the accuracy of training and validating data of VGG16 for
skin cancer images. The x-axis represents the number of epochs, while the y-axis

DOI: 10.4236/jilsa.2023.153005 72 Journal of Intelligent Learning Systems and Applications

 A. M. Ibrahim et al.

Figure 4. The accuracy of training and validating data of VGG16 for skin cancer images.

represents the accuracy percentage. The blue line represents the accuracy of
training data, while the orange line represents the accuracy of validating data. At
the beginning of training, both lines start at a low accuracy percentage. However,
as the number of epochs increases, both lines gradually increase in accuracy. The
training data achieves a higher accuracy percentage than the validating data
throughout most of the epochs. Towards the end of training, both lines start to
plateau and converge towards an accuracy percentage of around 84.242%. This
indicates that VGG16 has learned to accurately classify skin cancer images with a
high degree of precision. Overall, in this Figure 4 demonstrates that VGG16 is
an effective model for classifying skin cancer images with high accuracy on both
training and validating datasets.

4. Conclusions
Overall, transfer learning by VGG16 for skin cancer classification involves using
a pre-trained model to extract features from images of skin lesions and then
training a new output layer to make predictions about whether or not those le-
sions are malignant or benign.
To sum up, using transfer learning with VGG16 for skin cancer classification
can help us achieve high accuracy with relatively little data and avoid overfitting.
This makes it an attractive option for researchers and practitioners in the field of
dermatology who are looking to develop accurate and reliable methods for di-
agnosing skin cancer.

Conflicts of Interest
The authors declare no conflicts of interest regarding the publication of this pa-
per.

References
[1] Ali, M.S., Miah, M.S., Haque, J., Rahman, M.M. and Islam, M.K. (2021) An En-
hanced Technique of Skin Cancer Classification Using Deep Convolutional Neural

DOI: 10.4236/jilsa.2023.153005 73 Journal of Intelligent Learning Systems and Applications

A. M. Ibrahim et al.

Network with Transfer Learning Models. Machine Learning with Applications, 5,

Article ID: 100036. https://doi.org/10.1016/j.mlwa.2021.100036
[2] Carcagnì, P., Leo, M., Cuna, A., Mazzeo, P.L., Spagnolo, P., Celeste, G., et al. (2019)
Classification of Skin Lesions by Combining Multilevel Learnings in a DenseNet
Architecture. In: Ricci, E., Rota Bulò, S., Snoek, C., Lanz, O., Messelodi, S. and Sebe,
N., Eds., Image Analysis and Processing—ICIAP 2019, Springer, Cham, 335-344.
https://doi.org/10.1007/978-3-030-30642-7_30
[3] Dorj, U.O., Lee, K.K., Choi, J.Y. and Lee, M. (2018) The Skin Cancer Classification
Using Deep Convolutional Neural Network. Multimedia Tools and Applications,
77, 9909-9924. https://doi.org/10.1007/s11042-018-5714-1
[4] Pacheco, A.G. and Krohling, R.A. (2019) Recent Advances in Deep Learning Ap-
plied to Skin Cancer Detection. arXiv: 1912.03280.
[5] Ashraf, R., Afzal, S., Rehman, A.U., Gul, S., Baber, J., Bakhtyar, M., Mehmood, I.,
Song, O.Y. and Maqsood, M. (2020) Region-of-Interest Based Transfer Learning
Assisted Framework for Skin Cancer Detection. IEEE Access, 8, 147858-147871.
https://doi.org/10.1109/ACCESS.2020.3014701
[6] Byrd, A.L., Belkaid, Y. and Segre, J.A. (2018) The Human Skin Microbiome. Nature
Reviews Microbiology, 16, 143-155. https://doi.org/10.1038/nrmicro.2017.157
[7] Jeong, M.K., Lu, J.C., Huo, X., Vidakovic, B. and Chen, D. (2006) Wavelet-Based
Data Reduction Techniques for Process Fault Detection. Technometrics, 48, 26-40.
https://doi.org/10.1198/004017005000000553
[8] Namey, E., Guest, G., Thairu, L. and Johnson, L. (2008) Data Reduction Techniques
for Large Qualitative Data Sets. Handbook for Team-Based Qualitative Research, 2,
137-161.
[9] Mahbod, A., Schaefer, G., Wang, C., Ecker, R. and Ellinge, I. (2019) Skin Lesion
Classification Using Hybrid Deep Neural Networks. ICASSP 2019—2019 IEEE In-
ternational Conference on Acoustics, Speech and Signal Processing, Brighton, 12-17
May 2019, 1229-1233. https://doi.org/10.1109/ICASSP.2019.8683352
[10] Sagar, A. and Dheeba, J. (2020) Convolutional Neural Networks for Classifying Me-
lanoma Images. bioRxiv.
https://www.biorxiv.org/content/10.1101/2020.05.22.110973v2
[11] Shaikh, J., Khan, R., Ingle, Y. and Shaikh, N. (2022) Skin Cancer Detection: A Re-
view Using AI Techniques. International Journal of Health Sciences, 6, 14339-
14346.
[12] Althubiti, S.A., Alenezi, F., Shitharth, S., Sangeetha, K. and Simha Reddy, C.V.
(2022) Circuit Manufacturing Defect Detection Using VGG16 Convolutional Neur-
al Networks. Wireless Communications and Mobile Computing, 2022, Article ID:
1070405. https://doi.org/10.1155/2022/1070405
[13] Kalouche, S. (2016) Vision-Based Classification of Skin Cancer Using Deep Learning.
https://www.semanticscholar.org/paper/Vision-Based-Classification-of-Skin-Cance
r-using-Kalouche/b57ba909756462d812dc20fca157b3972bc1f533
[14] Bisla, D., Choromanska, A., Stein, J.A., Polsky, D. and Berman, R. (2019) Towards
Automated Melanoma Detection with Deep Learning: Data Purification and Aug-
mentation. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recogni-
tion Workshops (CVPRW), Long Beach, 16-17 June 2019, 2720-2728.
http://arxiv.org/abs/1902.06061
[15] Farag, A., Lu, L., Roth, H.R., Liu, J., Turkbey, E. and Summers, R.M. (2017) A Bot-
tom-Up Approach for Pancreas Segmentation Using Cascaded Superpixels and
(Deep) Image Patch Labeling. IEEE Transactions on Image Processing, 26, 386-399.

DOI: 10.4236/jilsa.2023.153005 74 Journal of Intelligent Learning Systems and Applications

 A. M. Ibrahim et al.

https://doi.org/10.1109/TIP.2016.2624198
[16] Le, D.N., Le, H.X., Ngo, L.T. and Ngo, H.T. (2020) Transfer Learning with Class-
weighted and Focal Loss Function for Automatic Skin Cancer Classification. arXiv:
2009.05977.
[17] Jafari, M.H., Karimi, N., Nasr-Esfahani, E., Samavi, S., Soroushmehr, S.M.R., Ward,
K., et al. (2016) Skin Lesion Segmentation in Clinical Images Using Deep Learning.
2016 23rd International Conference on Pattern Recognition, Cancun, 4-8 December
2016, 337-342. https://doi.org/10.1109/ICPR.2016.7899656
[18] Tschandl, P., et al. (2019) Comparison of the Accuracy of Human Readers versus
Machine-Learning Algorithms for Pigmented Skin Lesion Classification: An Open,
Web-Based, International, Diagnostic Study. The Lancet Oncology, 20, 938-947.
https://doi.org/10.1016/S1470-2045(19)30333-X

DOI: 10.4236/jilsa.2023.153005 75 Journal of Intelligent Learning Systems and Applications