Модель прогнозування захворювання на основі підходу конкатенації для стійкої системи охорони здоров’я
In the present world, due to many factors like environmental changes, food styles, and living habits, human health is constantly affected by different diseases, which causes a huge amount of data to be managed in health care. Some diseases become life-threatening if they are not cured at the startin...
Gespeichert in:
| Datum: | 2023 |
|---|---|
| Hauptverfasser: | , , |
| Format: | Artikel |
| Sprache: | Englisch |
| Veröffentlicht: |
The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"
2023
|
| Schlagworte: | |
| Online Zugang: | https://journal.iasa.kpi.ua/article/view/267594 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Назва журналу: | System research and information technologies |
| Завантажити файл: |  |
Institution
System research and information technologies| _version_ | 1866302880634372096 |
|---|---|
| author | Tharageswari, Kamaraj Mohana Sundaram, Natarajan Santhosh, Rajendran |
| author_facet | Tharageswari, Kamaraj Mohana Sundaram, Natarajan Santhosh, Rajendran |
| author_sort | Tharageswari, Kamaraj |
| baseUrl_str | http://journal.iasa.kpi.ua/oai |
| collection | OJS |
| datestamp_date | 2023-11-07T22:19:24Z |
| description | In the present world, due to many factors like environmental changes, food styles, and living habits, human health is constantly affected by different diseases, which causes a huge amount of data to be managed in health care. Some diseases become life-threatening if they are not cured at the starting stage. Thus, it is a complex task for the healthcare system to design a well-trained disease prediction model for accurately identifying diseases. Deep learning models are the most widely used in disease prediction research, but their performance is inferior to conventional models. In order to overcome this issue, this work introduces the concatenation of Inception V3 and Xception deep learning convolutional neural network models. The proposed model extracts the main features and produces the prediction result more accurately than traditional predictive models. This work analyses the performance of the proposed model in terms of accuracy, precision, recall, and f1-score. It compares the proposed model to existing techniques such as Stacked Denoising Auto-Encoder (SDAE), Logistic Regression (LR), MLP, MLP with attention mechanism (MLP-A), Support Vector Machine (SVM), Multi Neural Network (MNN), and Hybrid Convolutional Neural Network (CNN)-Random Forest (RF). |
| doi_str_mv | 10.20535/SRIT.2308-8893.2023.3.06 |
| first_indexed | 2025-07-17T10:28:00Z |
| format | Article |
| fulltext |
K. Tharageswari, N. Mohana Sundaram, R. Santhosh, 2023
Системні дослідження та інформаційні технології, 2023, № 3 81
TIДC
ТЕОРЕТИЧНІ ТА ПРИКЛАДНІ ПРОБЛЕМИ
ІНТЕЛЕКТУАЛЬНИХ СИСТЕМ ПІДТРИМАННЯ
ПРИЙНЯТТЯ РІШЕНЬ
UDC 62-50
DOI: 10.20535/SRIT.2308-8893.2023.3.06
A CONCATENATION APPROACH-BASED DISEASE
PREDICTION MODEL
FOR SUSTAINABLE HEALTH CARE SYSTEM
K. THARAGESWARI, N. MOHANA SUNDARAM, R. SANTHOSH
Abstract. In the present world, due to many factors like environmental changes,
food styles, and living habits, human health is constantly affected by different dis-
eases, which causes a huge amount of data to be managed in health care. Some dis-
eases become life-threatening if they are not cured at the starting stage. Thus, it is a
complex task for the healthcare system to design a well-trained disease prediction
model for accurately identifying diseases. Deep learning models are the most widely
used in disease prediction research, but their performance is inferior to conventional
models. In order to overcome this issue, this work introduces the concatenation of
Inception V3 and Xception deep learning convolutional neural network models. The
proposed model extracts the main features and produces the prediction result more
accurately than traditional predictive models. This work analyses the performance of
the proposed model in terms of accuracy, precision, recall, and f1-score. It compares
the proposed model to existing techniques such as Stacked Denoising Auto-Encoder
(SDAE), Logistic Regression (LR), MLP, MLP with attention mechanism (MLP-A),
Support Vector Machine (SVM), Multi Neural Network (MNN), and Hybrid Convo-
lutional Neural Network (CNN)-Random Forest (RF).
Keywords: feature extraction, disease prediction, deep learning, Inception V3,
Xception.
INTRODUCTION
The World Health Organization defines the health care system as the organization
of people that is mainly constructed to maintain, restore, and monitor the health
details of the public. The health system improves people’s health by providing
personal care given by hospitals and doctors. As a result, the primary goal of the
healthcare system is to keep people healthy by detecting diseases early and treat-
ing them appropriately. Obtaining an efficient health care system provides bene-
fits to maintaining people’s health.
In recent years, the development in the medical field has cured patients of
various diseases, but still, people are affected by some diseases due to their un-
predictable nature, and it causes a severe life-threatening problem for people. The
early prediction of those diseases saves the lives of many people. In the health
care system, doctors who use computer-aided diagnoses to quickly treat different
diseases need to be able to recognize, analyze, and classify data [5]. Thus, the
K. Tharageswari, N. Mohana Sundaram, R. Santhosh
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 82
healthcare system requires a well-trained disease predictive model to accurately
predict diseases. Based on the predicted disease, doctors can provide the correct
treatment and maintain the patient’s health. So, this proposed work designs a sys-
tem that can efficiently determine the condition to predict the disease based on the
given input data about the patient’s health.
The accuracy of the prediction result depends on the technique used in train-
ing the model, and the training depends on the sample data used for training. So,
the sufficiency of the data sample is very important in training the model to get a
better prediction result. In order to overcome the deficiency of sample data, the
transfer learning technique is used in the deep learning model. Deep learning is a
subtype of the machine learning approach. The rapid growth of deep learning in
various fields proposes various representative techniques. Because of the better
ability to learn the features of input data, deep learning has progressively replaced
traditional machine learning techniques [6]. Deep learning has the capacity to de-
termine features automatically from a given dataset for each specific application.
Transfer Learning is a kind of deep learning technique that uses pre-trained
knowledge from the past to train the new model and deploy the trained features of
a large dataset into a small dataset. Hence, using this pre-trained transfer learning
technique in the deep learning model will reduce the time taken for the classifica-
tion and prediction of disease [5]. However, the deep learning technique used by
the conventional model has several advantages, but the accuracy performance is
not better. So, the main motive of this research work is to improve prediction ac-
curacy by modifying the state-of-the-art method by concatenating networks. This
research work uses a concatenation of two different pretrained Convolutional
Neural Network (CNN) models called Inception V3 and Xception.
The Inception model consists of multiple convolution filters of various sizes,
and hence it can improve the adaptability of the network and extract more copious
features of different scales. Simultaneously, by using the Network in Network
model, the Inception model can significantly reduce the parameters of the model.
Hence, the network can minimize the number of convolution filters as much as
possible without losing model feature representation, thus minimizing the com-
plexity of the model [3]. Whereas, in the exception model, the term “exception”
refers to an extreme. The exception model takes the rule of the Inception model to
an extreme, so it provides the added advantage of feature extraction in a pointwise
manner also. The concatenation of two different models makes the proposed system
stronger due to the capability of multiple feature extraction [16]. The concatenation
model reduces the complexity and provides accurate prediction results based on
the given input data, which could be sustainable and reliable for health care.
The research contribution is summarized as follows:
Presented a sustainable healthcare prediction system using a concatena-
tion of Inception V3 and Xception to predict the disease based on input data.
Presented an intense experimental analysis to validate the performance of
the proposed model using standard benchmark data.
Presented a comparative analysis of the proposed model with existing
prediction models such as MLP, Logistic regression (LR), stacked denoising auto-
encoder (SDAE) and hybrid Convolutional Neural Network (CNN) – Random
Forest (RF) for performance validation.
The further discussions are arranged in the following order: Section two
comprises a detailed literature review and section three presents the proposed
model, experimental analysis and its results comparisons are presented in section
four and the conclusion is presented in the last section.
A concatenation approach-based disease prediction model for sustainable health care system
Системні дослідження та інформаційні технології, 2023, № 3 83
RELATED WORKS
In this section, research related to the existing disease prediction system was
enumerated. Various techniques, feature advantages, and disadvantages are re-
viewed for investigation, and finally, the limitations are discussed to structure the
research motivation. The healthcare system is rapidly becoming a necessary tool
for offering a well-rounded chance to meet the requirements of public health. In-
tegration of healthcare with a recommended system to meet the needs of older and
chronically diseased people was reported in [1]. Deep learning is a subfield of
machine learning that is broadly used in healthcare systems for the classification,
identification, and prediction of clinical data. A feature selection algorithm-based
prediction model reported in [3] uses a combination of the FCMIM-SVM ap-
proach to detect heart disease. The features for the prediction model are obtained
using a fast-conditional mutual information feature selection algorithm, and local
learning procedures are followed to remove the redundant and irrelevant features.
Machine learning techniques have been broadly adopted in various fields, particu-
larly in medical diagnosis. Multi-feature extraction using Inception-V3 and
Densnet-201 is achieved to predict the brain tumor [4].
A feature selection algorithm based on Chronic Obstructive Pulmonary
Disease prediction is reported in [6] uses an instance-based and feature-based
transfer learning Balanced Probability Distribution (BPD) and cross-domain
feature filtering algorithm. Various feature selection algorithms like the AdaBoost
algorithm, TCA algorithm, and Multi-Task Learning (MTL) algorithm are
compared with the BPD algorithm, and the result demonstrates the superior
prediction performance of the BPD algorithm. A comparative analysis of machine
learning techniques for disease prediction reported in [8] employs decision trees,
K-nearest neighbor, and logistic regression algorithms to predict kidney disease.
From the findings, it demonstrates that the decision tree approach and logistic
regression give better performance in predicting kidney disease. The article [5]
investigates the viability and effectiveness of various machine learning algorithms
like REP Tree, Random Tree, Linear Regression, M5P Tree, Naive Bayes, J48,
and JRIP to predict cardiovascular disease. And from the analysis, the result
demonstrates the superior prediction performance of the Random Tree
approach.
The computer-aided screening method reported in [17] utilizes a comparison
of 13 pre-trained CNN models like AlexNet, GoogleNet, VGG16, VGG19, etc.
and three deep learning based classifiers, namely K-Nearest Neighbor, Support
Vector Machine, and Naive Bayes for the analysis of Covid19 X-Ray and CT
Scan images. The result demonstrates that VGG19 with SVM classifier provides
superior performance to other methods. A novel Leaf GAN (Generative
Adversarial Network) method reported in [19] utilizes a data augmentation
method to identify disease-affected grape leaves. Initially, for training the GAN
model, four types of grape leaf disease images are generated using an image
generator model, and secondly, to identify original and duplicate images, an
image discriminator is used. Finally, a deep regret gradient penalty method is
employed for the stabilization of the GAN model.
A combination of convolutional neural networks with recursive neural net-
works reported in [20] utilizes an automatic prediction method for identifying and
categorizing the different kinds of blood cells. Due to a better understanding of
the features of blood cell images, accurate prediction and classification of blood
K. Tharageswari, N. Mohana Sundaram, R. Santhosh
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 84
cells are attained. The segmentation of breast masses in mammograms reported in
[21] utilizes a comparison of various kinds of deep learning models for the seg-
mentation and classification of breast lesions. The result demonstrates that the
VGG19 and ResNet50V2 models perform better in the classification of breast
lesions than other models.
The identification of the Citrus Disease Severity reported in [19] trains and
compares six different types of deep learning models to predict the Citrus Disease
Severity. From the analysis, it demonstrates that GAN-based data augmentation
with the Inception V3 model provides superior performance. A disease prediction
model reported in [2] utilizes a multi-stage model by the combination of co-
clustering and supervised machine learning methods to predict the intravenous
immunoglobulin resistance in Kawasaki disease. Initially, co-clustering is used to
cluster the missing data pattern blocks. Secondly, the selection of data features is
obtained by group lasso. Finally, the prediction of immunoglobulin resistance in a
patient is obtained using an explainable boosting method. The machine learning
approach reported in the article [7] utilizes a cloud-centric IoT system for the
prediction of skin disease. This research mainly focuses on the evaluation of six
deep learning models, namely VGG16, Inception, Xception, MobileNet,
ResNet50, and DenseNet161. Based on this evaluation, the article created a two-
phase classification process by the Targeted Ensemble Machine Classification
Model (TEMCM).
Apart from the classification of images, CNN are widely used for the
segmentation of images. The segmentation model reported in [20] utilized a CNN
along with an optimization technique to segment the various types of land in the
Amazon. A deep learning model reported in [8] utilizes the fused outputs of
ResNet50, Xception, and DenseNet in the FC layer of the super learner model for
classifying the type of vehicle. From the above survey, it is observed that the
performance of the prediction model depends on the selection of a suitable
learning algorithm and classifiers. Various types of machine learning algorithms
are mostly used in the research. From the above literature review, it is observed
that the Dense Net and ResNet perform well. However, in some of the articles,
concatenation of convolutional neural network models was utilized, but the
performance can be improved further using novel architectures. In order to extend
the performance of the concatenation-based CNN prediction system, a SoftMax
discriminator algorithm is used along with the concatenation model, and it is
discussed in the following section.
PROPOSED WORK
The proposed predictive model contains a concatenation of Inception V3 and
Xception CNN (Fig. 1). The proposed work consists of three stages. The first
stage involves Normalizing of the given medical input data. The second stage in-
volves features extraction from the input-data using concatenation approach. The
third stage involves the prediction of output using soft max discriminant classi-
fier.
Data Normalizing Stage. This stage starts from the collection of medical
datasets. Since the non-standardized input data’s take more time for learning
process, the data in the dataset are subjected to Normalizing procedure. Based on
the type of input data standardization will be applied to every input data to have
the same dimension/size.
A concatenation approach-based disease prediction model for sustainable health care system
Системні дослідження та інформаційні технології, 2023, № 3 85
Feature Extraction from the Concatenation Model. Feature extraction is
an essential step to be done before predicting the output, which is used to
determine knowledge from datasets and it mainly focus on the selection of
representative features from the given dataset, which will make the classifier to
easily diffirentiate between the various classes. Additionally, it also provide
benefit by reducing the computational time for training the classifier. A well-
organized feature extraction method will improve the accuracy of predicted
output. So, in this work, for attaining better feature extraction parallel deep
feature extraction technique based on transfer learning method is applied to the
concatenation of Inception V3 and Xception CNN models. The Normalized input
data is given to the Xception and Inception V3 CNN model for feature extraction.
To improve the quality of resultant feature map, the feature extracted from the
two models are combined. Xception model convolutes the input data by both
depth wise and pointwise manner (Fig. 2) and Inception model convolutes the
input data by depth wise manner (Fig. 3).
Prediction of output using Soft-max Discriminant Classifier (SDC). The
correct prediction of the given input data is done by the process of classification.
The task of the classifier is to approximate a feature map function from input data
to discrete output data. The important function of SDC is to identify the particular
class to which the testing data belongs to. This can be done by using the calcula-
tion of weighing distance between the test data and train data. In the proposed
work, the SDC is used as a binary classifier since it needs only two classes, where
class 1 represents abnormal status and the class 2 represents the normal status. The
following Figs. 2 and 3 show the architecture of Inception V3 Model and Xception
model respectively.
Consider the input medical dataset 1 2{ , , , }nD D D D . The normalization
of the input data is given by
)(min)(max
)(min
)(
ii
i
DD
DD
yxDf
,
where iD defines the input medical dataset of ,D x and y represents the constant of
normalizer.This standardized input datas are further given as input to the training
Fig. 1. Architecture of Proposed Concatenation Model
K. Tharageswari, N. Mohana Sundaram, R. Santhosh
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 86
and testing in the model. The standardized input data from the given medical
dataset is given as input to both the Xception and Inception V3 model. The feature
map expression of 1g from the pre-trained Inception V3 model is given
)),(|()( 1111111 bvgccriq i .
Output Channels
Fig. 2. Architecture of Xception
Fig. 3. Architecture of Inception V3
A concatenation approach-based disease prediction model for sustainable health care system
Системні дослідження та інформаційні технології, 2023, № 3 87
The feature map expression of 2g from the pre-trained Xception model is
given by
2 2 2 2 2 2 2( |( ( )),) iiq r c c g v b ,
where 1 v and 2v are the weight vector and 1b and 2b are the bias vector of the
Inception V3 and Xception model. The final feature map expression of G from
the concatenation model is given by
)),(;| ()( BVGcCRiQ i ,
where 1 2Q q q , 1 2R r r , 1cC + 2c , 1G g + 2g , 21 vvV and
B= 1 2B b b . The final feature map output obtained from the concatenation
network is given as input to the FC layer of the proposed mode ,which flattens
the input into a fixed length vector form. Next, the output of FC layer is given as
input to the drop out layer which is used for regularizing the data by avoiding
overfitting in the network. Finally, the output “s” from drop out layer is given to
the SDC for the classification of data.
Training and Testing of Data’s using SDC. Consider the training data set
be fe
m KSSSSS },,,,{ 321 is determined from the m distinct classes:
m
m
fem
f
mmm
m KSSSSS } .,,,,{ 321 represents totally mf data from the
mth class, where ffi
m
i
1
.
Consider the test data from the drop out layer as as 1es K For represent-
ing the test data “m” class data’s are used and by which a minimum reconstruc-
tion error is attained. The distance in between the m class data and the testing data
helps SDC for classifying the input data. The expression of SDC will be given by
the following equations:
i
sTsl maxarg)( ;
)||||(explogargmax)( 2
1
i
j
f
j
sssl
i
,
where i
ST , ( )l s defines the distance between the ith class and the test data, from
this s can be identified. defines the penalty parameter, which can be used when
0 . If “s” corresponds to the ith class, then s and i
js will have the same fea-
tures and hence
2
i
js s will goes to zero. This indirectly indicates that the fea-
tures of the test data match with the features of the ith class. Therefore i
sT helps
to attain the higher value asymptotically and hence i
sT is maximized. By this
way, SDC learns to predict the corresponding class of the given input data.
The prediction output of the SDC will be represented using binary classifica-
tion as follows:
K. Tharageswari, N. Mohana Sundaram, R. Santhosh
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 88
.0 if 2
,0if 1
s
s
Tclass
Tclass
ZOutput
Where P defines the decision condition of the SDC, which is predefined dur-
ing the training period of SDC. Class 1 defines the abnormal status and class 2
defines the normal status. The decision condition will get change based on the
type of disease.
Pseudocode for the proposed Concatenation of Inception V3 and Xcep-
tion Model
Input : D = {D1,D2,D3,….,Dn}
0 if 2
0 if 1
:
s
s
Tclass
Tclass
ZOutput
Begin
Initialize data normalization for the given input f (D)
Give the normalized data to both Inception V3 and Xception Pre-trained
Model
Obtain the feature map of Inception V3 model as 1q
Obtain the feature map of Xception model as 2q
Concatenate both pre-trained model and obtain final feature map as Q
Flatten the final feature map using FC layer
Regularize the FC output data using drop out layer
Give the regularized data to SDC and obtain the classification
Obtain the classification of SDC from i
sTSl maxarg)(
If 0 sT
SDC predicts the output as abnormal status
else
SDC predicts the output as normal status
End if
End
RESULTS AND DISCUSSION
The proposed Concatenation of Inception V3-Xception prediction model is
experimentally validated using MatLab 14.1 installed in an intel i3 processor
2.20 GHZ frequency with 8 GB memory. In order to obtain the accurate
prediction of disease, the metrices such as accuracy, precision, recall, and f1-score
are evaluated in the proposed work The first data set is collected from the
Cardiology Department of Chinese PLA General Hospital [25]. The feature
includes demographics, vital signs, lab tests, echocardiography, comorbidities,
length of stay, and medications. Total 105 features are obtained from each patient
and a total of 736 patient data are used in the dataset. The second data set is
collected from the UCI machine learning repository [26]. The dataset includes 76
features which include demographics, vital signs, cholesterol, echocardiography,
and medications. The benefit of using this dataset is that it allows investigating
the classification with multiple features. The simulation parameters used in the
proposed model is listed in Table 1.
A concatenation approach-based disease prediction model for sustainable health care system
Системні дослідження та інформаційні технології, 2023, № 3 89
T a b l e 1 . Simulation parameters
No Parameter Range/Value
1 Normalizer constant for input x=0.1, y=0.7
2 Learning rate 0.9
3 Number of epochs 500
TP means the actual medical data containing abnormal value is correctly
predicted as abnormal patient, FN means the actual medical data containing
abnormal value is incorrectly predicted as Normal patient, FP means the actual
medical data containing Normal value is incorrectly predicted as abnormal patient
and TN means the actual medical data containing Normal value is correctly
predicted as Normal data. The performance of the proposed work is determined in
terms of the following metrices:
FNTP
TP
Recall
;
FPTP
TP
Precision
;
Accuracy
FNFPTNTP
TNTP
;
F1-score
FNFPTP
TP
2
2
.
The Performance metrics of the proposed model is presented in the form of
the following Table 2.
T a b l e 2 . Performance metrics of proposed model
No Performance metrics Dataset 1 Dataset 2
1 Recall 0.993 0.991
2 Precision 0.990 0.987
3 F1-score 0.991 0.914
4 Accuracy 0.985 0.983
Further the performance of the proposed concatenation model has been
compared with existing techniques evaluated in Chen et al. [21] research work for
dataset 1 and Sudarshan et al. [26] research work for dataset 2. For the data set 1,
techniques like stacked denoising auto-encoder (SDAE), logistic regression (LR),
MLP, MLP with attention mechanism (MLP-A) and Multi neural networks
(MNN) are used to compare with proposed model. For the dataset 2 techniques
like support vector machine (SVM), logistic regression (LR), Random Forest
(RF), swarm artificial neural network (S-ANN) and multi neural networks are
used to compared with proposed concatenation model.
Figs. 4 and 5 presents the precision analysis of proposed model and
conventional models for dataset 1 and dataset 2 respectively. From the results, it
is clear that the proposed concatenation model exhibits maximum precision which
indicates the classification performance of proposed concatenation model has
been increased due to the multi feature selection and processing using soft-max
Discriminant classifier. Similarly, for dataset 2 the maximum performance is
K. Tharageswari, N. Mohana Sundaram, R. Santhosh
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 90
obtained by the proposed concatenation model whereas conventional methods
obtain minimum precision values compared to proposed concatenation model.
The average precision value attained by the proposed concatenation model for
dataset 1 is 0.990 and for dataset 2 the obtained precision is 0.987 which is much
better than the conventional methods.
The recall metrics of the proposed model and conventional models for
dataset 1 and dataset 2 has been presented in Figs. 4 and 5 respectively.
Results shows that the maximum recall obtained by the proposed model for both
datasets. Though the performance of MNN is much better than other conventional
techniques however it is lesser than the CNN-RF model and proposed
Algorithms
Fig. 4. Recall analysis for dataset 1
Algorithms
Fig. 5. Recall analysis for dataset 2
A concatenation approach-based disease prediction model for sustainable health care system
Системні дослідження та інформаційні технології, 2023, № 3 91
concatenation model. The average recall value obtained by the proposed model
for dataset 1 is 0.993 and for dataset 2 the obtained recall value is 0.991. The F1-
score analysis for the proposed model and existing models are comparatively
presented in Figs. 6 and 7 for dataset 1 and dataset 2 respectively. Based on the
recall and precision values, the f1-score has been obtained and presented. From
the results it clear that the maximum score is obtained by the proposed
concatenation model when compared to other conventional techniques. The
average f1-score obtained by the proposed concatenation model for dataset 1 is
0.991 and 0.914 for dataset 2.
The accuracy of the proposed concatenation model and conventional models
are comparatively analyzed and depicted in Figs. 8 and 9 for dataset 1 and dataset
2 respectively. It can be analyzed from the results; the maximum accuracy is
determined by the proposed concatenation model for both datasets whereas the
performances of conventional models are lesser than the proposed model accu-
Algorithms
Fig. 6. F1-score analysis for dataset 1
Algorithms
Fig. 7. F1-score analysis for dataset 2
K. Tharageswari, N. Mohana Sundaram, R. Santhosh
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 92
racy values. The maximum accuracy obtained by the proposed model is 0.985 for
dataset 1 and 0.983 for dataset 2. The accuracy obtained by MNN for dataset 1 is
0.966 and dataset 2 is 0.968 which is nearly 2% lesser than the proposed concate-
nation model. The accuracy obtained by CNN-RF for dataset 1 is 0.973 and data-
set 2 is 0.978 which is also nearly 1% lesser than the proposed concatenation
model. The multi feature selection using concatenation model and prediction us-
ing SDC increases the prediction accuracy of the proposed model. Whereas con-
ventional model performs less due to the improper feature selection and classifi-
cation process.
Table 3 shows the performance comparative analysis of proposed concatena-
tion model and conventional models in terms of accuracy, recall and precision.
The average values from the results of dataset 1 and dataset 2 are presented in the
tabulation. It can be observed from the results the performance of proposed con-
catenation model is much better than the conventional techniques. Thus, it is clear
that the proposed concatenation model can be used for predicting disease in health
cares to attain sustainable development.
Algorithms
Fig. 9. Accuracy analysis for dataset 2
Algorithms
Fig. 8. Recall analysis for dataset 2
A concatenation approach-based disease prediction model for sustainable health care system
Системні дослідження та інформаційні технології, 2023, № 3 93
T a b l e 3 . Performance comparative analysis
No Method Accuracy Precision Recall
1 Stacked denoising auto-encoder (SDAE) 0.623 0.670 0.782
2 Logistic regression (LR) 0.655 0.700 0.792
3 MLP 0.651 0.692 0.799
4 MLP with attention mechanism 0.667 0.710 0.795
5 SVM 0.87 0.85 0.85
6 Logistic regression 0.86 0.84 0.85
7 Random forest 0.89 0.88 0.88
8 Swarm-ANN 0.957 0.952 0.952
9 MNN 0.966 0.962 0.97
10 CNN-RF 0.973 0.982 0.987
11 Proposed Inception V3 – Xception -SDC 0.985 0.988 0.992
CONCLUSION
A concatenation model for disease prediction in healthcare system is presented in
this research work using InceptionV3-Xception model and Soft-max Discriminant
Classifier. The proposed architecture utilizes the multi-features extracted from
concatenation model and classify the data using Soft-max Discriminant Classifier.
The novelty in the architecture enhances the classification performance of data
analysis system compared to conventional CNN model. Standard healthcare data-
sets are used for experimentation and verified through performance metrics like
accuracy, recall, precision and f1-score.To demonstrate the better performance,
conventional techniques like stacked denoising auto-encoder (SDAE), logistic
regression (LR), MLP, MLP with attention mechanism (MLP-A), support vector
machine (SVM), Random Forest (RF), swarm artificial neural network (S-ANN)
,multi neural networks and Convolutional Neural Network-Random forest (CNN-
RF) are compared with proposed concatenation model. Experimental results de-
picts that the performance of proposed model is much better than the conventional
approaches. However, the performance of proposed concatenation model has sev-
eral benefits, the prediction result is possible only for binary classes whether the
data is normal or abnormal, that is considered as a minor limitation of this work.
Further this research work can be extended using multi classification to identify
the particular stage of the disease.
REFERENCES
1. Adekunle O. Afolabi and Pekka Toivanen, “Integration of Recommendation Systems
into Connected Health for Effective Management of Chronic Diseases,” IEEE Ac-
cess, vol. 7, pp. 49201– 49211, 2019.
2. Haolin Wang , Zhilin Huang, Danfeng Zhang, Johan Arief, Tiewei Lyu, and Jie
Tian,“Integrating Co-Clustering and Interpretable Machine Learning for the Predic-
tion of Intravenous Immunoglobulin Resistance in Kawasaki Disease,” IEEE Access,
vol. 8, pp. 97064–97071, 2020.
3. Jian Ping Li, Amin Ul Haq, Salah Ud Din, Jalaluddin Khan, Asif Khan, and Abdus
Saboor, “Heart Disease Identification Method Using Machine Learning Classifica-
tion in E-Healthcare,” IEEE Access, vol. 6, pp. 107562–107582, 2020.
4. Neelum Noreen, Sellappan Palaniappan, Abdul Qayyum, Iftikhar Ahmad, Mu-
hammad Imran, and Muhammad Shoaib, “A Deep Learning Model Based on
Concatenation Approach for the Diagnosis of Brain Tumor,” IEEE Access, vol. 8,
pp. 55135–55144, 2020.
K. Tharageswari, N. Mohana Sundaram, R. Santhosh
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 94
5. Rajkumar Gangappa Nadakinamani et al., “Clinical Data Analysis for Prediction of
Cardiovascular Disease Using Machine Learning Techniques,” Computational Intel-
ligence and Neuroscience, pp.1–13, 2022.
6. Qian Wang, Hong Wang, Lutong Wang, and Fengping Yu, “Diagnosis of Chronic
Obstructive Pulmonary Disease Based on Transfer Learning,” IEEE Access, vol. 8,
pp. 47370–47383, 2020.
7. Lodewijk Brand, Kai Nichols, Hua Wang, Li Shen, and Heng Huang, “Joint Multi-
Modal Longitudinal Regression and Classification for Alzheimer’s Disease Predic-
tion,” IEEE Transactions on Medical Imaging, vol. 14, pp. 1–10, 2019.
8. Gazi Mohammed Ifraz, Muhammad Hasnath Rashid, Tahia Tazin, Sami Bourouis,
and Mohammad Monirujjaman Khan, “Comparative Analysis for Prediction of Kid-
ney Disease Using Intelligent Machine Learning Methods,” Computational and
Mathematical Methods in Medicine, pp.1–10, 2021.
9. Hong Qing Yu and Stephan Reiff-Marganiec, “Targeted Ensemble Machine Classi-
fication Approach for Supporting IoT Enabled Skin Disease Detection,” IEEE Ac-
cess, vol. 8, pp. 50244–50252, 2021.
10. Mohamed A. Hedeya, Ahmad H. Eid, and Rehab F. Abdel-Kader, “A Super-Learner
Ensemble of Deep Networks for Vehicle-Type Classification,” IEEE Access, vol. 8,
pp. 98266–98280, 2020.
11. Cuiping Shi, Ruiyang Xia, and Liguo Wang, “A Novel Multi-Branch Channel Ex-
pansion Network for Garbage Image Classification,” IEEE Access, vol. 8,
pp. 154437–154452, 2020.
12. Aqsa Rahim, Yawar Rasheed, Farooque Azam, Muhammad Waseem Anwar, Mu-
hammad Abdul Rahim, and Abdul Wahab Muzaffar, “An Integrated Machine Learn-
ing Framework for Effective Prediction of Cardiovascular Diseases,” IEEE Access,
vol. 9, pp. 106575–106588, 2021.
13. Bin Liu, Cheng Tan, Shuqin Li, Jinrong He, and Hongyan Wang, “A Data Augmen-
tation Method Based on Generative Adversarial Networks for Grape Leaf Disease
Identification,” IEEE Access, vol. 8, pp. 102188–102198, 2020.
14. Rohit Bharti, Aditya Khamparia, Mohammad Shabaz, Gaurav Dhiman, Sagar Pande,
and Parneet Singh, “Prediction of Heart Disease Using a Combination of Machine
Learning and Deep Learning,” Computational Intelligence and Neuroscience,
pp. 1–11, 2021.
15. Muhammad Imran Razzak, Muhammad Imran, and Guandong Xu, “Efficient Brain
Tumor Segmentation with Multiscale Two-Pathway-Group Conventional Neural
Networks,” IEEE Journal of Biomedical and Health Informatics, pp.1–10, 2018.
16. Hania H. Faraga, Lamiaa A.A. Saidb, Mohamed R.M. Rizka, and Magdy Abd
ElAzim Ahmedc, “Hyperparameters optimization for ResNet and Xception in the
purpose of diagnosingCOVID-19,” Journal of Intelligent & Fuzzy Systems,
pp. 3555–3571, 2021.
17. Prabira Kumar Sethya, Santi Kumari Beherab, Komma Anithac, Chanki Pandeyd,
and M.R. Khand, “Computer aid screening of COVID-19 usingX-ray and CT scan
images:An inner comparison,” Journal of X-Ray Science and Technology, pp. 197–210.
18. Cheng Wang et al., “Pulmonary Image classification based on Inception-V3 Transfer
Learning Model,” IEEE Access, vol. 7, pp. 146533–146541, 2019.
19. Qingmao Zeng, Xinhui Ma, Baoping Cheng, Erxun Zhou, and Wei Pang, “GANs-
Based Data Augmentation for Citrus Disease Severity Detection Using Deep Learn-
ing,” IEEE Access, vol. 8, pp. 172882–172891, 2020.
20. Gaobo Liang, Huichao Hong, Weifang Xie, and Lixin Zheng, “Combining Convolu-
tional Neural Network With Recursive Neural Network for Blood Cell Image Classi-
fication,” IEEE Access, vol. 6, pp. 36188–36197, 2018.
21. Chen Huang , Xiaochen Wang, Jiannong Cao, Shihui Wang, and Yan Zhang, “HCF:
A Hybrid CNN Framework for Behavior Detection of Distracted Drivers,” IEEE Ac-
cess, vol. 8, pp. 109335–109349, 2020.
22. Joel Parente De Oliveira, Marly Guimarães Fernandes Costa, and Cícero Costa
Filho, “Methodology of Data Fusion Using Deep Learning for Semantic Segmenta-
tion of Land Types in the Amazon,” IEEE Access, vol. 8, pp. 187864–187875, 2020.
A concatenation approach-based disease prediction model for sustainable health care system
Системні дослідження та інформаційні технології, 2023, № 3 95
23. Andrés Anaya-Isaza, Leonel Mera-Jiménez, Johan Manuel Cabrera-Chavarro, Lo-
rena Guachi-Guachi, Diego Peluffo-Ordóñez, and Jorge Ivan Rios-Patiño, “Com-
parison of Current Deep Convolutional Neural Networks for the Segmentation of
Breast Masses in Mammograms,” IEEE Access, vol. 9, pp. 152206–152225, 2021.
24. Sin-Ae Lee, Hyun Chin Cho, and Hyun-Chong Cho, “A Novel Approach for In-
creased Convolutional Neural Network Performance in Gastric-Cancer Classification
Using Endoscopic Images,” IEEE Access, vol. 9, pp. 51847–51854, 2021.
25. Yan Zhao, Baoqiang Ma, Pengbo Jiang, Debin Zeng, Xuetong Wang, and Shuyu Li,
“Prediction of Alzheimer’s Disease Progression with Multi-Information Generative
Adversarial Network,” IEEE Journal of Biomedical and Health Informatics,vol. 25,
pp. 711–719, 2020.
26. Sudarshan Nandy, Mainak Adhikari, Venki Balasubramanian, Varun G. Menon,
Xingwang Li, and Muhammad Zakarya, “An intelligent heart disease prediction sys-
tem based on swarm artificial neural network,” Neural Computing and Applications,
pp. 1–15, 2021.
Received 27.11.2022
INFORMATION ON THE ARTICLE
K. Tharageswari, Karpagam Academy of Higher Education, Coimbatore, India, e-mail:
ktharageswari4@gmail.com
Dr. N. Mohana Sundaram, Karpagam Academy of Higher Education, Coimbatore, India
Dr. R. Santhosh, Karpagam Academy of Higher Education, Coimbatore, India, e-mail:
santhoshrd@gmail.com
МОДЕЛЬ ПРОГНОЗУВАННЯ ЗАХВОРЮВАННЯ НА ОСНОВІ ПІДХОДУ
КОНКАТЕНАЦІЇ ДЛЯ СТІЙКОЇ СИСТЕМИ ОХОРОНИ ЗДОРОВ’Я / К. Тара-
гесварі, Н. Мохана Сундарам, Р. Сантош
Анотація. У сучасному світі внаслідок багатьох факторів, таких як зміни на-
вколишнього середовища, стилі харчування та життєві звички, на здоров’я
людей постійно впливають різні захворювання, що призводить до того, що в
системі охорони здоров’я потрібно керувати величезною кількістю даних. Де-
які захворювання створюють небезпеку для життя, якщо їх не вилікувати на
початковій стадії. Для системи охорони здоров’я це робить складним завдан-
ням розробити добре навчену модель прогнозування захворювань для точної їх
ідентифікації. Моделі глибокого навчання найбільш широко використовують-
ся в дослідженнях прогнозування захворювань, але їх продуктивність поступа-
ється звичайним моделям. Щоб вирішити цю проблему, у роботі подано кон-
катенацію моделей згорткових нейронних мереж глибокого навчання Inception
V3 і Xception. Запропонована модель виділяє основні ознаки та створює ре-
зультат прогнозу точніше, ніж інші традиційні моделі прогнозування. У роботі
аналізується продуктивність запропонованої моделі з точки зору точності,
прецизійності, запам’ятовування та F1-міра, порівнюються моделі з існуючими
методами, такими як стековий автоматичний кодувальник (SDAE), логістична
регресія (LR), MLP, MLP з механізмом уваги (MLP-A), опорна векторна ма-
шина (SVM), мультинейронна мережа (MNN), гібридна згорткова нейронна
мережа (CNN), випадковий ліс (RF).
Ключові слова: вилучення функцій, прогнозування захворювань, глибоке
навчання, Inception V3, Xception.
|
| id | journaliasakpiua-article-267594 |
| institution | System research and information technologies |
| keywords_txt_mv | keywords |
| language | English |
| last_indexed | 2025-07-17T10:28:00Z |
| publishDate | 2023 |
| publisher | The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" |
| record_format | ojs |
| resource_txt_mv | journaliasakpiua/97/dce3742024c8db41dfb3b3647c66e397.pdf |
| spelling | journaliasakpiua-article-2675942023-11-07T22:19:24Z A concatenation approach-based disease prediction model for sustainable health care system Модель прогнозування захворювання на основі підходу конкатенації для стійкої системи охорони здоров’я Tharageswari, Kamaraj Mohana Sundaram, Natarajan Santhosh, Rajendran вилучення функцій прогнозування захворювань глибоке навчання Inception V3 Xception feature extraction disease prediction deep learning Inception V3 Xception In the present world, due to many factors like environmental changes, food styles, and living habits, human health is constantly affected by different diseases, which causes a huge amount of data to be managed in health care. Some diseases become life-threatening if they are not cured at the starting stage. Thus, it is a complex task for the healthcare system to design a well-trained disease prediction model for accurately identifying diseases. Deep learning models are the most widely used in disease prediction research, but their performance is inferior to conventional models. In order to overcome this issue, this work introduces the concatenation of Inception V3 and Xception deep learning convolutional neural network models. The proposed model extracts the main features and produces the prediction result more accurately than traditional predictive models. This work analyses the performance of the proposed model in terms of accuracy, precision, recall, and f1-score. It compares the proposed model to existing techniques such as Stacked Denoising Auto-Encoder (SDAE), Logistic Regression (LR), MLP, MLP with attention mechanism (MLP-A), Support Vector Machine (SVM), Multi Neural Network (MNN), and Hybrid Convolutional Neural Network (CNN)-Random Forest (RF). У сучасному світі внаслідок багатьох факторів, таких як зміни навколишнього середовища, стилі харчування та життєві звички, на здоров’я людей постійно впливають різні захворювання, що призводить до того, що в системі охорони здоров’я потрібно керувати величезною кількістю даних. Деякі захворювання створюють небезпеку для життя, якщо їх не вилікувати на початковій стадії. Для системи охорони здоров’я це робить складним завданням розробити добре навчену модель прогнозування захворювань для точної їх ідентифікації. Моделі глибокого навчання найбільш широко використовуються в дослідженнях прогнозування захворювань, але їх продуктивність поступається звичайним моделям. Щоб вирішити цю проблему, у роботі подано конкатенацію моделей згорткових нейронних мереж глибокого навчання Inception V3 і Xception. Запропонована модель виділяє основні ознаки та створює результат прогнозу точніше, ніж інші традиційні моделі прогнозування. У роботі аналізується продуктивність запропонованої моделі з точки зору точності, прецизійності, запам’ятовування та F1-міра, порівнюються моделі з існуючими методами, такими як стековий автоматичний кодувальник (SDAE), логістична регресія (LR), MLP, MLP з механізмом уваги (MLP-A), опорна векторна машина (SVM), мультинейронна мережа (MNN), гібридна згорткова нейронна мережа (CNN), випадковий ліс (RF). The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 2023-09-29 Article Article application/pdf https://journal.iasa.kpi.ua/article/view/267594 10.20535/SRIT.2308-8893.2023.3.06 System research and information technologies; No. 3 (2023); 81-95 Системные исследования и информационные технологии; № 3 (2023); 81-95 Системні дослідження та інформаційні технології; № 3 (2023); 81-95 2308-8893 1681-6048 en https://journal.iasa.kpi.ua/article/view/267594/283970 |
| spellingShingle | вилучення функцій прогнозування захворювань глибоке навчання Inception V3 Xception Tharageswari, Kamaraj Mohana Sundaram, Natarajan Santhosh, Rajendran Модель прогнозування захворювання на основі підходу конкатенації для стійкої системи охорони здоров’я |
| title | Модель прогнозування захворювання на основі підходу конкатенації для стійкої системи охорони здоров’я |
| title_alt | A concatenation approach-based disease prediction model for sustainable health care system |
| title_full | Модель прогнозування захворювання на основі підходу конкатенації для стійкої системи охорони здоров’я |
| title_fullStr | Модель прогнозування захворювання на основі підходу конкатенації для стійкої системи охорони здоров’я |
| title_full_unstemmed | Модель прогнозування захворювання на основі підходу конкатенації для стійкої системи охорони здоров’я |
| title_short | Модель прогнозування захворювання на основі підходу конкатенації для стійкої системи охорони здоров’я |
| title_sort | модель прогнозування захворювання на основі підходу конкатенації для стійкої системи охорони здоров’я |
| topic | вилучення функцій прогнозування захворювань глибоке навчання Inception V3 Xception |
| topic_facet | вилучення функцій прогнозування захворювань глибоке навчання Inception V3 Xception feature extraction disease prediction deep learning Inception V3 Xception |
| url | https://journal.iasa.kpi.ua/article/view/267594 |
| work_keys_str_mv | AT tharageswarikamaraj aconcatenationapproachbaseddiseasepredictionmodelforsustainablehealthcaresystem AT mohanasundaramnatarajan aconcatenationapproachbaseddiseasepredictionmodelforsustainablehealthcaresystem AT santhoshrajendran aconcatenationapproachbaseddiseasepredictionmodelforsustainablehealthcaresystem AT tharageswarikamaraj modelʹprognozuvannâzahvorûvannânaosnovípídhodukonkatenacíídlâstíjkoísistemiohoronizdorovâ AT mohanasundaramnatarajan modelʹprognozuvannâzahvorûvannânaosnovípídhodukonkatenacíídlâstíjkoísistemiohoronizdorovâ AT santhoshrajendran modelʹprognozuvannâzahvorûvannânaosnovípídhodukonkatenacíídlâstíjkoísistemiohoronizdorovâ AT tharageswarikamaraj concatenationapproachbaseddiseasepredictionmodelforsustainablehealthcaresystem AT mohanasundaramnatarajan concatenationapproachbaseddiseasepredictionmodelforsustainablehealthcaresystem AT santhoshrajendran concatenationapproachbaseddiseasepredictionmodelforsustainablehealthcaresystem |