A Novel Hybrid Deep Learning Model for Detecting Breast Cancer

Abstract

Cancer is a deformity of the body cells that grow out of control and spread to other parts of body. According to the American Cancer Society, early identification of cancer resulted in a 99% chance of survival in the localized stage. In this article, a benchmark cancer dataset is considered for the purpose of investigation. After the preprocessing of data, we attempt to predict cancer type using our proposed deep learning model named as CNN-Soft-T that combines the feature selection (FS) and classification. FS is introduced using histogram technique for anomaly detection and dimension reduction. We have compared the performance of this model to other cutting-edge machine learning approaches including Logistic Regression (LR), Naive Bayes (NB), Support Vector Machine (SVM), Random Forest (RF), Neural Network (NN) and Decision Tree (DT) in order to demonstrate its utility. We have experimentally exhibited the exceptional empirical effectiveness and biological applicability of our suggested hybrid model for cancer diagnostics and medication development.

Full text

2007 Afr. J. Biomed. Res. Vol. 27, No.4s (November) 2024
Avishek Banerjee
et al.
https://africanjournalofbiomedicalresearch.com/index.php/AJBR
Afr. J. Biomed. Res. Vol. 27(4s) (November 2024); 2007-2016
Research Article
A Novel Hybrid Deep Learning Model for Detecting Breast
Cancer
Avishek Banerjee
1
, Debasis Chakraborty
2
, Sambit S. Mondal
3
, Shemim
Begum
4
, Bikas Mondal
5
, Sanghamitra Layek
6
and
Santana Das
7
1
Department of Information Technology,
Asansol Engineering College, Asansol-713305
2,3
Department of Computer Science and Engineering (IoT, CS, BCT),
Asansol Engineering College,
Asansol-713305
4
Department of Computer Science and Engineering, Government College of Engineering and
Textile Technology, Berhampore – 742101
5,6
Department of Electronics and Computer Science,
Narula Institute of Technology.
7
Department of Electronics and Computer Science, Guru Nanak Institute of Technology
Email:
1
avishekbanerji@gmail.com,
2
debasisju67@gmail.com,
3
sambitsmondal@gmail.com,
4
sh em im _b eg um @yahoo.com,
5
bikas.mondal@nit.ac.in,
6
sanghamitra.layek@nit.ac.in,
7
santana.das@gnit.ac.in
Abstract
:
Cancer is a deformity of the body cells that grow out of control and spread to other parts of body.
According to the American Cancer Society, early identification of cancer resulted in a 99% chance of survival in the
localized stage. In this article, a benchmark cancer dataset is considered for the purpose of investigation. After the
preprocessing of data, we attempt to predict cancer type using our proposed deep learning model named as CNN-Soft-T
that combines the feature selection (FS) and classification. FS is introduced using histogram technique for anomaly
detection and dimension reduction. We have compared the performance of this model to other cutting-edge machine
learning approaches including Logistic Regression (LR), Naive Bayes (NB), Support Vector Machine (SVM), Random
Forest (RF), Neural Network (NN) and Decision Tree (DT) in order to demonstrate its utility. We have
experimentally exhibited the exceptional empirical effectiveness and biological applicability of our suggested hybrid
model for cancer diagnostics and medication development.
Keywords:
Breast cancer, Convolutional neural network, Data cleaning, Average Accuracy, Hybrid deep learning
model.
*Author for correspondence:
avishekbanerji@gmail.com
Received: 10/10/24 Accepted: 19/11/24
DOI: https://doi.org/10.53555/AJBR.v27i4S.3987
© 2024 The Author(s).
This article has been published under the terms of Creative Commons Attribution-Noncommercial 4.0 International
License (CC BY-NC 4.0), which permits noncommercial unrestricted use, distribution, and reproduction in any
medium, provided that the following statement is provided. “This article has been published in the African Journal of
Biomedical Research”
1.
INTRODUCTION
Cancer means uncontrolled and unregulated growth of
neoplastic cells. It spreads very quickly and
vigorously in different organs in our body leading
to fatality. Genetic disorder, any type of gene
mutation i.e., change in genome sequences are
responsible factors for developing cancers. Accurate
detection of different tumor types can help in
providing better treatment and toxicity minimization
on the patients. Radiology and imaging play a vital role
for detecting abnormalities or stages of disorder. There
are different types cancer such as Carcinoma,

A Novel Hybrid Deep Learning Model for Detecting Breast Cancer
2008 Afr. J. Biomed. Res. Vol. 27, No.4s (November) 2024
Avishek Banerjee
et al.
lymphoma, sarcoma, leukemia, and many more. The
rapid advancement of ML and deep learning (DL) in
particular continues to fuel the interest of medical
community in leveraging these methods to improve the
accuracy of cancer screening. Some of the ML
techniques are developed for the detection of cancer
including leukemia, prostate tumor, mixed-lineage
leukemias (MLL) Diffuse large B- cell lymphomas
(DLBCL) and Childhood acute lymphoblastic leukemia
(ALLGSE412)[1], [2], [3] and many more. Breast
cancer is clinically detected such as physical
examination, mammography, and biopsy. Fine needle
aspiration (FNA), core needle biopsy, surgical biopsy,
or lymph node biopsy are the four methods that can
be used to perform a biopsy [4]. Although surgical
biopsy leads to 100% accuracy, it is however expensive,
time-consuming, and painful. Using anomalies like
masses and/or micro calcifications, mammography can
identify breast cancer. Mammographic images [5] are
very useful for detecting radio-diagnosing disorders
and anomalies. DL and ML are therefore being used
increasingly in clinical cancer research in order to
reduce cost and time.
In this study a breast cancer dataset downloaded from
the website [6] has been considered. The dataset
consisting of 30 different features of breast cells and
two classes namely benign, or malignant is considered
for investigation. While benign means it won’t infect
neighbouring tissues or move to other parts of the
body, malignancy is substantially more deadly because
it spreads throughout the body. The accuracy of
diagnosing breast cancer and predicting whether it is
malignant/cancerous or benign/non-cancerous is
evaluated using a variety of ML models, including LR,
NB, SVM, RF, DT, and the proposed hybrid DL model.
The dataset is first mined in order to reduce
abnormalities. Data abnormalities, redundant data, and
data with missing values are all minimized during the
data cleaning process. Following the discovery of
anomalies, three features are found having significant
anomalies. Subsequently, six ML models such as LR,
DT, NN, NB, SVM, and RF are applied to estimate the
accuracy level by considering 30 and 27 features
separately. Empirical analysis shows that all the models
outperform when the number of features is less.
The remainder of the article is divided into the
following sections: Section 2 provides a literature
review. Different approaches, including the proposed
model, are described in section 3. The result analysis is
shown in part 4, and the conclusion is presented in
section 5.
2. LITERATURE REVIEW
L. Liu [7] employed
the LR approach to diagnose breast cancer in 2018.
Data related to biopsy cells from women with typical
breast lumps were analyzed using an LR approach to
gauge the efficacy of ML for cancer diagnosis. The LR
approach was used in this study to classify the datasets
for breast cancer diagnosis using the Sklearn ML
toolbox. in 2019. Khairunnahar, L. et al. [8] identified
malignant and benign tissue using LR in 2019. For the
waiting factor, the sigmoid function was selected. This
waiting element was shown to be present in the data
collection and type of optimization strategy that
considerably increased accuracy. High accuracy
classifiers suggested a lower risk of breast cancer. RF
algorithm for precise breast cancer prediction was
proposed by B. Dai et al. in 2018[9]. H. Chougrad et al.
[10] proposed the idea of cancer screening. The
radiologists were given instructions on how to find
mass lesions using mammography, breast biopsies,
Computer assisted diagnosis (CAD), and CNN.
Transfer learning also proved highly successful for
medical imaging. Using CNN and RBF-based SVM,
M. Alkhaleefah et al. (2018) [11] presented a method
for classifying breast cancer in mammograms.
A reliable diagnosis technique, Convolutional Neural
Network for breast cancer classification (CNNI-BCC),
was proposed by Ting, F. F. et al. in [5]. In this
experiment, tumors were successfully classified into
benign and malignant kinds. It was really helpful to
medical professionals to categorize mammographic
images using this technique. A model that classifies
breast tumors using DT-ML and K-Nearest Neighbours
(KNN) techniques was reported by Rajaguru, H. et al.
[12]. The comparison of the study’s findings revealed
that the KNN classifier outperformed the decision- tree
classifier in terms of diagnosing breast cancer. In
another approach, SVMs and deep CNNs were used by
D. A. Ragab et al. [13] in 2019 as a method for
identifying breast cancer. Using ensemble Hoeffding
trees and NB, Alhayali et al. [14] reported a successful
method for diagnosing breast cancer in 2020. An
Improved Random Forest-Rule Extraction (IRFRE)
approach was proposed by Wang, S. et. al. [15].
A superior Multi Objective Evolutionary Algorithm
(MOEA) was applied for precision and interpretability.
The developed method was evaluated using three
datasets, including Wisconsin Diagnostic Breast
Cancer, Wisconsin Original Breast Cancer, and
Surveillance, Epidemiology, End-Result (SEER) Breast
Cancer. Variable importance measures (VIM) were
suggested as a method for diagnosing breast cancer in
2021 by Huang, Z. et al. [16]. The Hierarchical
Clustering Random Forest (HCRF) algorithm using
VIM was the most precise approach. Breast cancer was
examined in [17] utilizing the RF Classifier and Grey
Level Co-occurrence Matrix (GLCM). They suggested
integrating NSS (Neighbourhood Structural
Similarities) and GLCM. There were two different NSS
models: NSS1 and NSS2. RF was used to detect
healthy or malignant tissue.
Desai, M. et al. [18] announced the identification of
breast cancers in 2021. For a higher degree of accuracy,
CNN and Multilayer Perceptron Neural Network
(MLP) were utilised for malignancy identification.
Using this procedure, each operational network was
designed, diagnosed, and then classified before being
analysed. Patient-reported out- come measures
(PROMs) were proposed in 2022 by Coriddi, M. et al.
in [19] for identifying lymphedema linked to breast
cancer. Both the Upper Limb Lymphedema 27 scale

A Novel Hybrid Deep Learning Model for Detecting Breast Cancer
2009 Afr. J. Biomed. Res. Vol. 27, No.4s (November) 2024
Avishek Banerjee
et al.
(ULL27) and the Lymphedema Life Impact Scale
(LLIS) appeared to be quite specific for lymphedema
and able to distinguish it from symptoms brought on by
axillary lymph node dissection (ALND) alone. The
proposed results suggested that these threshold-based
questionnaires would be useful for identifying
lymphedema, potentially reducing the need for frequent
clinic visits and time-consuming evaluations. In order
to examine several approaches for multiple feature
selection and find the smallest set of features that might
reliably diagnose breast cancer as benign or malignant,
Khan, M. M. et al. suggested a novel CNN model in
2022 [20].
3. METHODOLOGY
A. Different existing machine learning models
Different ML algorithms are discussed if brief as
follows:
1) Logistic Regression (LR): LR [21] is a supervised
method of categorizing data. A decision threshold is
involved in a classification strategy, and the
classification problem itself impacts how the threshold
value is determined, which is significant in LR.
2) Naive Bayes Classifier (NB): The Naive Bayes[22]
probabilistic machine learning approach, which is used
for a range of classification issues, is built on the Bayes
Theorem. Despite recent developments in ML, it has
shown to be not only quick, accurate, and dependable
but also easy.
3) Support Vector Machine Algorithm:: Classification
and regression problems are resolved using one of the
most well-known supervised learning algorithms,
Support Vector Machine, or SVM[23]. But the ML
Classification problem makes extensive use of it. To
categorise test data points in n- dimensional space, the
SVM algorithm’s goal is to determine the best line or
decision boundary.
4) Random Forest: The supervised learning approach
includes Random Forest (RF) [24]. It may be used to
solve classification and regression-related ML
problems. It is based on the concept of ensemble
learning, a technique for combining several classifiers
to solve complex problems and improve model
performance.
5) Decision Tree: Decision tree algorithms are a subset
of supervised learning algorithms. The DT approach
can deal with both classification and regression
problems, unlike other supervised learning techniques.
A decision tree [25] is used to create a training model
that can be applied to forecast the class or value of the
target variable by learning fundamental choice rules
learned from prior data (training data).
B. Proposed CNNs with SoftMax function and a
temperature hyper-parameter (CNN-Soft-T)
Convolutional Neural Networks (CNN)[26] have
emerged as an effective DL technology for computer
vision problems. CNNs have achieved outstanding
success in a variety of disciplines, from picture
classification to object recognition and beyond, thanks
to their capacity to autonomously build hierarchical
representations from visual data. CNNs continue to
drive advances in computer vision and play an
important role in a wide range of real-world
applications.
CNNs have emerged as an effective DL technology for
computer vision problems. CNNs have achieved
outstanding success in a variety of disciplines, from
picture classification to object recognition and beyond,
thanks to their capacity to autonomously build
hierarchical representations from visual data. CNNs
continue to drive advances in computer vision and play
an important role in a wide range of real-world
applications.
The following definition can be used to describe the
mathematical model for CNN-Soft-T in the context of
breast cancer detection:
Algorithm
1. Dataset Preparation: Gather a labelled collection of
breast cancer samples, with each sample represented by
a unique set of characteristics. Divide the dataset into
two parts: training and testing.
2. Data Preprocessing: To achieve consistency,
normalize the feature values to an acceptable range. To
improve the training data, use any necessary feature
engineering techniques, such as scaling or
dimensionality reduction.
3. CNNs Architecture: Create a CNN architecture for
breast cancer diagnosis using feature-based data. The
CNN will be fed a 1D vector containing the
characteristics of each sample. Convolutional layers
can be used to extract local patterns and capture crucial
feature correlations. Include pooling layers to decrease
dimensionality and down sample features. To execute
the categorization task, add fully linked layers at the
end. After each layer, apply suitable activation
functions (e.g., ReLU).

A Novel Hybrid Deep Learning Model for Detecting Breast Cancer
2010 Afr. J. Biomed. Res. Vol. 27, No.4s (November) 2024
Avishek Banerjee
et al.
Fig 1. Block diagram of modified CNN
Fig. 2. Complete work flow of the technique

A Novel Hybrid Deep Learning Model for Detecting Breast Cancer
2011 Afr. J. Biomed. Res. Vol. 27, No.4s (November) 2024
Avishek Banerjee
et al.
4. Training: Create a CNN with random weights. Feed
the training samples into the network and use the
SoftMax function with a temperature hyperparameter to
determine the output probabilities. To calculate the
difference between predicted and true labels, use a loss
function such as cross-entropy. Use an optimization
approach (in this research the stochastic gradient
descent) to iteratively update the network weights while
minimizing the loss. To optimize the training process,
change the learning rate and other hyperparameters.
Rep till the network converges or a predetermined
number of epochs is achieved.
5. SoftMax-Temperature Hyperparameter:
Introduce a temperature hyperparameter (denoted as T)
to modify the SoftMax output probabilities.
Mathematically, given an input vector z = [z1, z2, ...,
zn], the SoftMax function computes the probability
󰇛󰇜 for each element zi as follows:
󰇛󰇜 = exp(zi) / (exp(z1) + exp(z2) + ... + exp(zn)) …(1)
Apply the SoftMax function to the logits (pre-SoftMax values) with the temperature adjustment: SoftMax (logits / T).
Higher values of T (e.g., T > 1) make the SoftMax distribution softer and more uniform, while lower values (e.g., T < 1)
amplify the differences between probabilities.
󰇛󰇜 = (/T)/(( + + ... + )/T) …(2)
Experiment with different temperature values to balance the model's sensitivity to subtle differences in the input.
To approximate softmax(z) using a Taylor series expansion, we can expand ez and keep a few terms.


󰇛󰇜󰇛󰇜

 …(3)
In the experiment the most of the values used in the different fields of the dataset is mixed integer and the range was in
between 0 up to 2 digits, +ve mixed integer and that is reason behind using approximate the CNN function.




󰇛󰇜



 ….(4)


󰇟󰇛󰇜󰇠
󰇣󰇡
󰇢󰇤
󰇣󰇡
󰇢󰇤
󰇟󰇡
󰇢󰇠
 …(5)
Therefore the Eqn 2 could be approximated as:
󰇛󰇜

󰇛󰇩󰇟󰇛󰇜󰇠






󰇪
󰇜󰇛 + + ... + )/T …(6)
Where x is the upper limit of the fields in the dataset.
6. Inference: Use the trained model to categorize fresh,
previously unknown breast cancer feature samples
during the inference phase. To determine the class
probabilities, use the SoftMax function with the
temperature hyperparameter of choice. If required,
assign the class label based on the greatest likelihood or
set a confidence threshold.
7. Evaluation: Utilized measures of accuracy to assess
the trained model's performance on the testing dataset.
If the performance is poor, fine-tune the
hyperparameters or model design. To guarantee
generalizability, use cross-validation or other rigorous
assessment approaches.
The paper proposed hybrid deep learning model i.e.,
CNN-Soft-T models. The main objective of the paper is
as follows.
• The model is the combination of Convolutional
Neural Network (CNN), Softmax function and the
temperature Hyperparameter.
• The CNN is used for handling many data viz. 27
features and 30 feature datasets with multiple class e.g.,
malignant, and benign in a multi layers network. This
CNN models are trained by batch gradient back
propagation method .
• The SoftMax function is used for its exponentiation,
which gives a non-negative curve and here it is being
divided by a positive temperature hyperparameter value
(T>1) which make the SoftMax distribution softer and
more uniform, while lower values e.g., T < 1 amplify
the differences between probabilities.
The complete process of the proposed model is
illustrated in Fig. 2
4. RESULT ANALYSIS
Investigation has been conducted using six different
algorithms of ML algorithms viz. LR, NB, SVM, RF,
DT, NN and one proposed novel deep learning
algorithm CNN-Soft-T. The predictive performance of
different algorithms are assessed using the following
metrics average accuracy and standard deviation. These
two metrics are obtained using 10 fold cross validation
accuracy. The steps in this process are as follows. First,
10 subgroups of equal size are selected at random from

A Novel Hybrid Deep Learning Model for Detecting Breast Cancer
2012 Afr. J. Biomed. Res. Vol. 27, No.4s (November) 2024
Avishek Banerjee
et al.
the training samples. The classifier is then trained ten
times, using the remaining nine subsets to train the
classifier while holding out one subset at a time. The
classifier is then put to the test on the subset that was
held out, and the classification accuracy is noted. To
get the average accuracy, the accuracies are averaged.
Python is used to implement all algorithms. First, the
dataset with 30 features is used to determine accuracy
and SD, and each ML model was ran fifty times to
produce the desired results. The accuracy (%) was
calculated for three scenarios: the average the worst
and the best scenarios. The dataset is then cleaned up to
remove anomalies, and 27 features are selected to
estimate accuracy and SD again. The expression
levels of the training and test sets for the top six
characteristics are shown in Fig. 3. The heatmaps,
which are arranged as a feature vs sample matrix, show
how effective CNN-Softmax-T’s chosen features are in
differentiating between classes. The left side of the
pictures shows the characteristics. The figure shows
that the chosen characteristics are differently expressed
in benign and malignant classes for both training and
test datasets.
Best case, worst case, and average case accuracy SD
for both 27 and 30 features are reported in Table I. It
can be observed from the table that the proposed model
(model 4) outperforms other existing models. Learning
rate, layers and epoch are illustrated in Fig. 4a for all
four models. For instance, layers [25,1,1] mean we
applied 25 input layers, 1 hidden layers and 1 output
layers. The learning rate hyperparameter controls how
quickly an algorithm updates or learns the values of a
parameter estimate. It is represented by the symbol and
in this research it is increasing in model 3 and 4. A
hyperparameter known as ”epoch” estimates how many
times the learner will run on the entire training dataset.
It is observed that in model 4, the epoch reaches to its
highest value. Best results are shown in bold face.
Moreover, our model exhibits the best average
accuracy with 27 features. In Fig. 4b, a boxplot
depicting the % accuracy of the various approaches is
shown for demonstration. The boxplot for CNN-
Softmax-T, which is located at position six from the
left in the picture, is clearly visible on the upper side of
the figure. This shows that CNN-Softmax-T generates
accuracy scores that are greater than those generated by
the other approaches. The method is suitable for the
classification of cancer. In Table II, best case and worst
case accuracies are reported for the proposed and
existing techniques. Best results are in shown in bold
face. It is clear from the table that out model has the
best empirical success.
Table I : Related work on different types of ML m odels, accuracy (%) a nd standard-deviation
(SD ) for breast cancer detection
Model
Name
Layered
model
Model
Worst
accuracy
with 27
features
Worst
accuracy
with 30
features
Best accuracy
with 27
features
Best accuracy
with 30
features
Avg. accuracy
with 27
features
Avg. accuracy ±
SD
with 30
features
LR
NA
91.83
90.65
97.39
95.37
94.32 ± 1.492
92.95 ± 1.218
NB
NA
94.49
93.61
96.61
95.49
95.48 ± 0.648
94.41 ± 0.503
SVM
NA
84.80
84.21
91.23
90.64
88.31±1.705
86.84 ± 1.680
RF
NA
93.61
92.36
96.49
95.84
95.02 ±0.809
93.95 ±1.134
DT
NA
91.67
89.51
95.81
95.87
93.57 ±1.178
92.56 ±1.847
CNN-Soft-
T
Model 1
62.28
62.28
62.28
62.28
NA
NA
Model 2
94.74
95.61
94.77
95.61
NA
NA
Model 3
95.61
93.86
95.61
93.86
NA
NA
Model 4
96.15
95.83
99.16
96.89
97.75 ±0.955
96.37 ±0.303
Table II : Comparison of accuracy among the best and worst case for proposed ML models and
existing literature [19, 20]
Sl No
Model Name
Best Case Accuracy %
(27 features)
Worst Case Accuracy %
(27 features)
1
LR
97.39
91.83
2
NB
96.61
94.49
3
SVM
91.23
84.80
4
RF
96.49
93.61
5
DT
95.81
91.67
6
CNN-Soft-T
99.16
96.15
7
Lymphedema Life Impact Scale (LLIS)
(Coriddi et. al., 2022) [19]
97
NA
8
Upper Limb Lymphedema-27 scale (ULL27S)
(Coriddi et. al., 2022) [19]
93
NA
9
CNN (Khan et. al., 2022) [20]
99
NA
10
LR (Khan et. al., 2022) [20]
96
NA
11
RF (Khan et. al., 2022) [20]
98
NA
12
SVM (Khan et. al., 2022) [20]
97
NA
13
Voting Classifier (VC) [20]
(Khan et. al., 2022)
97
NA
14
DT (Khan et. al., 2022) [20]
96
NA

A Novel Hybrid Deep Learning Model for Detecting Breast Cancer
2013 Afr. J. Biomed. Res. Vol. 27, No.4s (November) 2024
Avishek Banerjee
et al.
Mainly the research has been performed using six
different algorithms of ML algorithms viz. LR, NB,
SVM, RF, DT, NN and one proposed novel deep
learning algorithm CNN-Soft-T. Using those
algorithms, the main objective of the research was to
detect the accuracy % of those models in the dataset of
breast cancer. Python language has been used for
implementation purpose. The dataset with 30 features
used to find the accuracy and run fifty times for each
ML models to get the desired output. The Accuracy
(%) calculated for three cases viz. average case, worst
case and best case and standard deviation also. After
that the dataset has been cleaned up to reduce the
Anomalies and selected 27 features out of 30 features
and calculated Accuracy (%) for three cases viz.
average case, worst case, and best case and standard
deviation also. In 2022, Khan et. al. the researcher
achieved 96% classification accuracy with respect to
DT and 97% classification accuracy with respect to
SVM which is much greater than the research held by
the current research therefore the researchers decided to
perform 10-fold cross validation for those models.
In Table1, the researchers compared the accuracy
obtained through all 6 different models. It has been
observed that the proposed CNN-Soft-T model
produced best accuracy in case of dataset with 27
features i.e., 97.61 and in case of average case the
proposed CNN-Soft-T model produced better result in
both cases i.e., in case of dataset with 27 features and
dataset with 30 features.
In Table 2, Comparison of accuracy among best and
worst case for proposed ML models and existing
literature (Coriddi et. al., 2022 and Khan et. al., 2022)
has been depicted and the proposed CNN-Soft-T
model having has highest accuracy percentage than
other existing models.
Fig.3: Boxplot of best-case and worst-case accuracy for different ML and DL methods.
The above figure significantly depicting the boxplot of the proposed CNN-Softmax-T method is higher than all other
considered methods for this experiment.

A Novel Hybrid Deep Learning Model for Detecting Breast Cancer
2014 Afr. J. Biomed. Res. Vol. 27, No.4s (November) 2024
Avishek Banerjee
et al.
Fig. 3(A)
Fig. 3(B)
Fig. 3A,3B. Heatmaps represent the best six
features’ expression levels for 18 samples from the
training and test sets. The examples are adjusted to
make it simple to identify similarities within classes
and differences between classes. Each column
corresponds to a sample, whereas each row represents
a feature
(a)
(b)
Fig. 4. (a) Learning rate and layers of four models (b) Boxplot showing accuracies of different methods
The heatmaps of the expression levels of the top six
features selected by the CNN-Softmax-T method. Each
row represents a feature and each column corresponds
to a sample.
5. Conclusion
In this research, we address the issue of breast cancer
categorization by introducing a unique deep learning
model. This work is novel in two ways. To increase
accuracy, the approach might first choose the pertinent
feature subset. Second, we have incorporated additional
parameter T for further increase in efficiency.
Obtained result of our model is compared with the
existing techniques to demonstrate the superiority of

A Novel Hybrid Deep Learning Model for Detecting Breast Cancer
2015 Afr. J. Biomed. Res. Vol. 27, No.4s (November) 2024
Avishek Banerjee
et al.
our model. The researchers applied various existing
algorithms after necessary fine tuning of hyper-
parameters as well as the researchers proposed an
effective hybrid deep learning model i.e., CNN-
Softmax-K model which lead to a higher classification
in comparison to other equivalent deep learning models
from existing literature and significant improvement is
found in this work. Additionally, as part of our future
research, we plan to use hybrid algorithms to increase
this domain’s accuracy. The future researchers also can
apply other hybrid algorithms to obtain higher accuracy
percentage in this research field.
REFERENCES
1. Maulik, U., & Chakraborty D. (2014), Fuzzy
preference based feature selection and
semisupervised SVM for cancer classification.
IEEE Transactions on NanoBioscience. 13(2),
152-160.
2. Begum, S., Sarkar R., Chakraborty D., & Maulik
U, (2020) Identification of biomarker on biological
and gene expression data using fuzzy preference
based rough set. Journal of Intelligent Systems.
30(1), 131-141.
3. Begum, S., Sarkar R., Chakraborty D., Ghosh M.,
& Maulik U. (2021), Application of active learning
in DNA microarray data for cancerous gene
identification. Expert Systems with Applications.
177.
4. Khandezamin, Z., Naderan, M., & Rashti, M. J.
(2020). Detection and classification of breast
cancer using logistic regression feature selection
and GMDH classifier. Journal of Biomedical
Informatics, 111, 103591.
5. Ting, F. F., Tan Y. J., & Sim, K. S. (2019)
Convolutional neural network improvement for
breast cancer classification. Expert Systems with
Applications, 120, 103-115.
6. website: https://www.kaggle.com/datasets/yasserh/
breast-cancer-dataset.
7. Liu, L. (2018), Research on logistic regression
algorithm of breast cancer diagnose data by
machine learning. IEEE International Conference
on Robots Intelligent System (ICRIS),157-160.
8. Khairunnahar, L., Hasib, M. A., Rezanur, R. H.
B., Islam, M. R., &Hosain, M. K. (2019).
Classification of malignant and benign tissue with
logistic regression. Informatics in Medicine
Unlocked, 16, 100189.
9. Dai, B., Chen, R. C., Zhu, S. Z., & Zhang, W. W.
(2018, December). Using random forest algorithm
for breast cancer diagnosis. In 2018 International
Symposium on Computer, Consumer and Control
(IS3C) (pp. 449-452). IEEE.
10. Chougrad, H., Zouaki, H., & Alheyane, O. (2018).
Deep convolutional neural networks for breast
cancer screening. Computer methods and
programs in biomedicine, 157, 19-30.
11. Alkhaleefah, M., & Wu, C. C. (2018, October). A
hybrid CNN and RBF-based SVM approach for
breast cancer classification in mammograms.
In 2018 IEEE International Conference on
Systems, Man, and Cybernetics (SMC) (pp. 894-
899). IEEE.
12. Rajaguru, H., & SR, S. C., (2019). Analysis of
decision tree and k-nearest neighbor algorithm in
the classification of breast cancer. Asian Pacific
journal of cancer prevention: APJCP, 20(12),
3777.
13. Ragab, D. A., Sharkas, M., Marshall, S., & Ren, J.
(2019). Breast cancer detection using deep
convolutional neural networks and support vector
machines. PeerJ, 7, e6201.
14. Alhayali, R. A. I., Ahmed, M. A., Mohialden, Y.
M., & Ali, A. H. (2020). Efficient method for
breast cancer classification based on ensemble
hoffeding tree and naïve Bayes. Indonesian
Journal of Electrical Engineering and Computer
Science, 18(2), 1074-1080.
15. Wang, S., Wang, Y., Wang, D., Yin, Y., Wang,
Y., &Jin, Y. (2020). An improved random forest-
based rule extraction method for breast cancer
diagnosis. Applied Soft Computing, 86, 105941.
16. Huang, Z., & Chen, D. (2021). A breast cancer
diagnosis method based on VIM feature selection
and hierarchical clustering random forest
algorithm. IEEE Access, 10, 3284-3293.
17. Kumar, T. A., Rajakumar, G., & Samuel, T. A.
(2021). Analysis of breast cancer using grey level
co-occurrence matrix and random forest
classifier. International Journal of Biomedical
Engineering and Technology, 37(2), 176-184.
18. Desai, M., & Shah, M. (2021). An anatomization
on breast cancer detection and diagnosis
employing multi-layer perceptron neural network
(MLP) and Convolutional neural network
(CNN). Clinical eHealth, 4, 1-11.
19. Coriddi, M., Kim, L., McGrath, L., Encarnacion,
E., Brereton, N., Shen, Y., ... & Dayan, J. H.
(2022). Accuracy, sensitivity, and specificity of
the LLIS and ULL27 in detecting breast cancer-
related lymphedema. Annals of surgical
oncology, 29, 438-445.
20. Khan, M. M., Tazin, T., Zunaid Hussain, M.,
Mostakim, M., Rehman, T., Singh, S., ... &
Alomeir, O. (2022). Breast Tumor Detection
Using Robust and Efficient Machine Learning and
Convolutional Neural Network
Approaches. Computational Intelligence and
Neuroscience, 2022.
21. Shipe M. E.,S. Deppen A., Farjah F., Grogan L.
E.(2019), Developing prediction models for
clinical use using logistic regression: an overview.
Journal of Thoracic Disease, 11(Suppl 4), S574.
22. Salmi N. & Rustam Z.(2019), Naıve Bayes
classifier models for predicting the colon cancer, In
IOP conference series: materials science and
engineering, 546(5),05206-8.
23. Vapnik V. & Izmailov R. (2021), Reinforced SVM
method and memorization mechanisms, Pattern
Recognition, 119,10801-8.
24. Schonlau M. & Zou R. Y. (2020), The
random forest algorithm for statistical learning”,
The Stata Journal, 20(1),3-29.

A Novel Hybrid Deep Learning Model for Detecting Breast Cancer
2016 Afr. J. Biomed. Res. Vol. 27, No.4s (November) 2024
Avishek Banerjee
et al.
25. Charbuty B. & Abdulazeez A.(2021),
Classification based on decision tree algorithm
for machine learning, Journal of Applied Science
and Technology Trends, 2(01), 20-28.
26. Desai M. & Shah M.(2021), An anatomization on
breast cancer detection and diagnosis employing
multi-layer perceptron neural network (MLP) and
Convolutional neural network (CNN), Clinical
eHealth,4, 1-11.