GRD Journals- Global Research and Development Journal for Engineering | Volume 6 | Issue 5 | April 2021

ISSN- 2455-5703

Chronic Kidney Disease Stage Prediction in HIV

Infected Patient using Deep Learning
Dr. Sheshang Degadwala Dhairya Vyas
Associate Professor Managing Director
Department of Computer Engineering Shree Drashti Infotech LLP, Nizampura, Vadodara,
Gujarat, India
Sigma Institute of Engineering, Vadodara, Gujarat

Abstract
The CKD is the worldwide phenomenon with high morbidity and death rates. Chronic renal disease (CKD). Since the early stages
of the CKD do not have any symptoms, patients frequently struggle to recognize their condition. HIV-patients are most likely to
suffer from critically compromised kidney failure. Early diagnosis of CKD allows patients to get prompt medication to improve
the disease's development. The suggested CNN deep learning model for the organization of the CKD phases observed with HIV
is presented in this article. The credits of CKD patients are carried out on site. In the Chronic Kidney Disease phase predicted,
CNN is 99% accurate with the PCA model.
Keywords- Chronic Kidney Disease, Stage, Machine Learning, Deep Learning, Convolution Nural Network, Principle
Componend Analysis

I. INTRODUCTION
Chronic kidney disease is an increasing global health concern (CKD). This is an incurable illness linked to a rise in morbidity and
death, an increased risk for many other illnesses, including cardiac failure and higher costs for health care. More than two million
patients worldwide undergo dialysis or kidney transplants to remain alive, but this figure may only account for 10% of people
requiring survival therapy [2]. In only five rich nations, the bulk of the 2 million individuals who receive anti-kidney disease
therapy constitute 12 percent of the world's population. In contrast, in about 100 developed countries just 20% of the world's
population is handled, representing about half of the world's population. More than a million people die each year from unexplained
kidney disease because of the enormous financial cost of dialysis and renal transplant care in 112 low-income countries [5].
When this form of circumstance occurs, filters do not function properly in order to allow HIV to infect the cells of the
kidney, HIV may damage glomeruli (nephrons). If any drugs used to treat HIV are not closely controlled, the nephrons can damage
the kidney. Since several studies were carried out to identify CKD or not and to detect CKD steps. The relationship between CKD
and HIV is, however, novel.
The rate of incidence and prevalence of CKD in the context of HIV infection varies in all areas with significant differences
and on the same continent. Variety depends on a number of factors such as kidney function tests, CKD definition, genetic variation,
prevention program, access to a health care system and the implementation of integrated ART. The first obstacle to overcome is
proper kidney function tests because no methods have been used to measure the glomerular filtration rate (eGFR) certified in
PLWH.

Fig. 1: Causes of Chronic Kidney Disease [16]

Therefore, it is very important to detect, control, and control the disease early. It is necessary to predict the progression
of CKD with due accuracy due to its strong and subtle nature in the early stages, as well as the heterogeneity of the patient. CKD
is often described in stages of severity. Clinical decisions are influenced by the stage, whether the patient is progressing, and the
level of progression. Also, defining the stage of the disease is very important as it provides many indications that support the
determination of the necessary interventions and treatment.

All rights reserved by www.grdjournals.com 53

 Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning
(GRDJE/ Volume 6 / Issue 5 / 007)

Table 1: Chronic Kidney Disease Stages

Stages Explanation GFR
One Normal damage of kidney function >90%
Two Minor damage of kidney job 89-60%
Three a Minor to Modest damage 59-45%
Three b Modest to simple damage 44-30%
Four Simple damage of kidney meaning 29-15%
Five Kidney Stop Working <15%
Automated computer aided diagnose is process of getting stage information by input the patient data so according it will take
action.
Algorithms for the prediction and differentiation of machine learning in healthcare is used. In order to identify and predict
diabetic and predominant patients, Yu et al [2] have used the Support Vector Machine algorithm to indicate that SMV helps classify
patients with common disorders. M N. Amin et al.[4] have now been diagnosed by the use of the Support Vector Machine (SVM)
for the analysis of a full group of patients with anatomical magnetic resonance imaging (MRI) and the findings indicate that SVM
is an encouraging means of diagnosing Alzheimer's disease at an early stage. PRNN is more effective than other cardiovascular
prediction algorithms. E. Perumal et a cardiovascular disease prediction with the Algorithm for Probabilistic Neural Network,
Algorithm for Decision Tree and the Algorithm for Naïve Bayes and PRNN. R. Shinde et al. [8] conducted HBV-induced hepatic
cirrhosis predictors using the MLP and the findings indicate a satisfactory predictive result for liver disorders, in particular HBV
patients with heptic insufficient disease with the MLP separator. R. Shinde et al.

II. LITERATURE PAPERS

The thesis was conducted by Jack Edward, Corinne Isnard Bagnis, David M. Gracey et al (2020) The more regrettable prosperity
of chronic kidney infection/disorder is synonymous with PLWWHIV, higher terribleness and mortality, and is strongly linked to
CVD. If a person is in high risk of being chronically infected with the kidney or already has chronic kidney disease it should
possibly prevent nephrotoxic ART.
A analysis (2019) undertaken in Chinese by J. Qin, L. Chen, Y. Liu, C. Liu, C. Feng and B. Chen et al.[2], Chinese. The
CKD diagnostic approach enables data imputation and sample diagnosis. For non-institute functionality using the KNN adjective,
the integrated model may achieve ample consistency behind the Independent Loan. Open data research is mostly closed with only
400 models available due to contextual problems. Modeling approach. Likewise, the model's power could decline. In addition, the
model does not explore the reality of chronic kidney disease since only two categories of data checks are used (Chronic kidney
and Not Chronic kidney diseases).
E. H. A. Rady and A. S. Anwar et al. [3] On a clinical/lab dataset of 361 persistent kidney sickness patients, using PNN,
SVM, RBF, and MLP mining calculations. The results of the kept an eye on computations were stood out all together from sort
out which figuring conveyed the most exact results in gathering the reality period of Chronic kidney sickness. The Probabilistic
Neural Networks computation is the best figuring for specialists to use to diminish assurance and treatment botches, as shown by
this report. A. Al Imran, M. N. Amin, and F. T. Johora et al. [4] Experimental discoveries were analysed for two circumstances,
model presentations on real (imbalanced) data and model displays on oversampled (changed) data, using Logistic Regression,
Feedforward Neural Networks, and Large and Deep Learning. For both certifiable and oversampled data, feedforward neural
associations made the best execution, with a 0.99 F1-score, 0.97 Precision, 0.99 Recall, and 0.99 AUC score.
N. Chetty, K. S. Vaisla, and S. D. Sudarsan et al. [5] On the Chronicle kidney disease dataset quality evaluator and
arrangement models were utilized. Utilizing NB, SMO, and IBK Classifiers, bringing about better characterization precision on
the decreased dataset than the first dataset.
P. Arulanthu and E. Perumal et al. [6] To lessen the ascribes of the first dataset, use JRip, SMO, Naive Bayes, and IBK
Attributes evaluators, and analyse the after effects of the first and diminished datasets, as the classifier using the first dataset
characterization results functioned admirably, yet the proposed classifier didn't woks for Jring for multiple data rows.
K. Shankar, P. Manickam, G. Devika, and M. Ilayaraja et al. [7] To order Chronic kidney illness, specialists used a
streamlining model and a learning convention. The interaction chooses pertinent highlights of kidney information utilizing the Ant
Lion Optimization (ALO) strategy to pick the best highlights for grouping. To accomplish high arrangement exactness, accuracy,
F-measure, and affectability by utilizing DNN.
Using the following: Maurya, R. Wable, R. Shinde, R. Jadhav, S. John and Dakshayani, R. [8]; Patients can be guided by
the amount of potassium they have collected by using the levels of blood potassium to delay the movement of chronic kidney
infection.
The author describes the importance of highlight preference procedures to improve the precise and implementing
arrangements mechanisms, which are used to create and classify different techniques for determining or reducing their
dimensionality.

III. PROPOSED SYSTEM

As show in the below diagram of proposed CNN based deep learning system for CKD stage classification of HIV patient.

All rights reserved by www.grdjournals.com 54

 Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning
(GRDJE/ Volume 6 / Issue 5 / 007)

.
Fig. 2: Proposed System

A. CKD Dataset
The dataset was purchased from the University of California, Irvine Machine Learning Repository. The dataset includes 26
functions, including 19 numerical attributes and 7 categorical ones. There were 158 separate instances in this dataset. The highlights
include age, a blood pressure, explicit weight loss, egg whites, sugar, red platelets, discharge cell, bacteria, irregular blood glucose,
blood urea, serum creatinine, sodium, hemoglobin, volumes of the pressed cells, white platelets, red platelets, hypertension and
diagnosis.
Table 2: Dataset Colum of HIV CKD Patient
Sr no Data Type
1 Age Numerical
2 Gender Categorical
3 ethnicity Numerical
4 Blood Pressure Numerical
5 Specific Gravity Numerical
6 Albumin Numerical
7 Sugar Numerical
8 Red Blood Cells Numerical
9 Pus Cell Numerical
10 Pus Cell clumps Numerical
11 Bacteria Numerical
12 Blood Glucose Random Numerical
13 Blood Urea Numerical
14 Serum Creatinine Numerical
15 Sodium Numerical
16 Potassium Numerical
17 Hemoglobin Numerical
18 Packed Cell Volume Numerical
19 White Blood Cell Count Numerical
20 Red Blood Cell Count Numerical
21 Hypertension Numerical
22 Diabetes Mellitus Categorical
23 Coronary Artery Disease Categorical
24 Appetite Categorical
25 Pedal Edema Categorical
26 Anemia Categorical
27 Class Categorical

B. Noise Removal
Most standard Noise removal techniques is obtained because of the abuse of shortenings, data passage botches, copy records,
missing esteems. Inside this specific circumstance, one key research subject is the de-duplication issue which is the recognition
and expulsion of copy records from a database. The exploration challenge is that databases contain both correct and vague copies.
In the stage, the non-numerical data are cleared and procured the numerical dataset for proceeding with further.

1) Shortenings [3]
Data is not valuable in an array. Data is only valuable once information, insight or in other words knowledge is extracted from it
and is used to make decisions. In data shorting data is divided into equal size raw and column.

2) Data Passage Botches [5]

In this part data is process in badly manner so the data passage botches algorithm will remove bugs data.

All rights reserved by www.grdjournals.com 55

 Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning
(GRDJE/ Volume 6 / Issue 5 / 007)

3) Copy Records [2]

It is better to construct a single statement that removes all unwanted copies in one go. Before removing the duplicate records, you
must decide which instances you want to keep.

4) Missing Esteems [7]

Missing value imputation is one of the important tasks in machine learning especially in the cases where the data is small and there
is a need of using all available data. Therefore, it plays an important role in the classification performance of the models. As our
dataset is small and all the features except the output feature contain missing values, we have an obvious need for imputing missing
values. There are many missing value imputation techniques in machine learning, however, we have used the mean and mode
imputation in this research. The numerical features were imputed using mean and the categorical features were imputed using
mode.

C. eGFR Calculation
Mostly, the phases of ongoing kidney illness are controlled by the assessed glomerular filtration rate (eGFR). There are five phases,
with two sub-stages 3a and 3b in stage 3. Kidney work is normal at stage one. To analyses CKD, two examples should be required
in any event 90 days separated, with recorded qualities utilized too. Creatinine computation, sex, identity, and age all impact the
assessed Glomerular Filtration Rate (eGFR). Alteration of Diet in Renal Disease is quite possibly the most dependable techniques
for ascertaining eGFR.
Correct it and Calculate egfr with this equation.
eGFR=175 x (Creatinine/88.4)^ 1.154 x (Age) ^ 0.203 x (0.742 if female) x (1.210 if dark) (1)
Example we have value of Creatinine=4.1 Age=68 ethnicity=black Gender= Male then the value of eGFR= 175x (4.1/88.4)
^1.154x (68) ^0.203x1.210= 14.41 so stage is 5. As it is between 0 to 15.

D. Attributes Selection RFE

Recursive Feature Elimination, or RFE for short, is a popular Attributes selection algorithm. RFE is popular because it is easy to
configure and use and because it is effective at selecting those features (columns) in a training dataset that are more or most relevant
in predicting the target variable. There are two important configuration options when using RFE: the choice in the number of
features to select and the choice of the algorithm used to help choose features. Both of these hyperparameters can be explored,
although the performance of the method is not strongly dependent on these hyperparameters being configured well. For our data
28 data column after applying RFE it will be reduced to 14 data columns

E. Dimensition Reduction PCA

In this research use orthogonal transformation in this process, which is a statistical technique for transforming a collection of
potentially correlated values known as principal components into a single value (PCA). When using PCA as a feature selection
method, the basic concept is to choose variables based on the magnitude (from largest to smallest in absolute values) of their
coefficients (loadings). PCA aims to substitute pp (more or less correlated) variables with k<pk<p uncorrelated linear combinations
(projections) of the original variables, as you may remember. Let's set aside the question of how to choose the best kk for the
problem at hand. The value of the kk principal components is determined by their explained variance, and each variable contributes
to each component to varying degrees. Using the largest variance criteria is similar to function extraction, in which the principal
attribute is used as a new feature rather than the original variables. However, we can choose to keep only the first factor and pick
the j<pj<p variables with the highest absolute coefficient; the number jj may be determined by the proportion of the total number
of variables (e.g., keep only the top 10% of the pp variables) or by a fixed cut-off. (e.g., considering a threshold on the normalized
coefficients). For our data 28 data column after applying PCA it will be reduced to 13 data columns.

F. CNN Layer Description

1) Conv
In-depth study, the convolutional neural network is a group of deep neural networks, often used to analyse visual images. They are
also known for fixed or fixed space exchange networks that are inserted, based on the construction of shared weights and translation
signals.

Fig. 3: Convolution Example

All rights reserved by www.grdjournals.com 56

 Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning
(GRDJE/ Volume 6 / Issue 5 / 007)

2) MaxPool
Max Pool is a blending capacity that chooses the most noteworthy item in the element map district covered by a channel. In this
manner, the yield next the maximum pooling layer will be an element map that contains the best noticeable highlights of the older
component map.

Fig. 4: Pooling Example

3) Flatten
Flattening is changing over the information into a 1-dimensional cluster for contributing it to the following layer. We smooth the
returns of the convolutional layers to make a lone long segment vector. Additionally, it is related with the last course of action
model, which is known as a totally related layer.

Fig. 5: Flatten Example

4) Dense
The dense layer is a neural association layer that is related significantly, which infers each neuron in the dense layer gets
commitment from all neurons of its past layer. The dense layer is found to be the most ordinarily used layer in models. Far out, the
thick layer plays out a grid vector increase.

5) Dropout
Dropout can be utilized after convolutional layers (for example, Conv2D) and in the wake of pooling layers (for sample
MaxPooling2D). Regularly, dropout is just utilized after the pooling layers, however, this is only a harsh heuristic. For this
situation, dropout is applied to every component or cell inside the element maps.

Fig. 6: Dropout Example

6) FineTune
Fine-tuning an organization is a methodology dependent on the idea of move learning [1,3]. We begin preparing CNN to learn
highlights for a wide area with an order work focused at limiting blunder in that space.

All rights reserved by www.grdjournals.com 57

 Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning
(GRDJE/ Volume 6 / Issue 5 / 007)

Fig. 7: Fine-Tune Flow of Work

Fine-Tune is a process based on the concept of learning transfer. Here our proposed model is beginning to train CNN to
learn the characteristics of a broader domain with a segmentation function aimed at reducing error on that domain. After that, it
will replace the partition function and use the network and reduce the error in another domain. Under this setting, transfer network
features and parameters from a broad domain to a specific one. Classification function uses SoftMax classification in this network
learning.

IV. PARAMETERS, RESULTS AND ANALYSIS

In this part discuss about different parameters definition and also showcase results. At the end comparative analysis is done with
existing methods.
TP + TN
Acccuracy = (2)
TP + FP + TN + FN
TP
Precision = (3)
TP + FP
TP
Recall = (4)
TPFN
2∗TP
F1 − Score = (5)
2∗TP+FP+FN
Where, true positive (TP), true negative (TN), false positive (FP) and false negative (FN)
As visible in figure 8 CNN is train using all 24 attributes and accuracy report is plot for recall, precision and f1-score.

Fig. 8: Convolution Neural Network (CNN) Result

All rights reserved by www.grdjournals.com 58

 Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning
(GRDJE/ Volume 6 / Issue 5 / 007)

As visible in figure 9 CNN is train using all 13 attributes which is selected by PCA and accuracy report is plot for precision, recall,
and f1-score.

Fig. 9: Convolution Neural Network (CNN)+ PCA Result

Table 3: Result Analysis of HIV and non-HIV Datasets

Classifier Attributes Non-HIV Accuracy (%) HIV Accuracy (%)
SVM 14 89 93
KNN 14 99 97
DT 14 99 97
RF 14 98 95
Ada Boost 14 99 97
XgBoost 14 98 97
CNN 24 97.88 99
CNN-PCA 13 98.66 99
Above Table 3 shows Comparative study of Different Classifier of Machine Learning Approaches and Deep learning Approaches
using Attribute and accuracy parameter. Due to Blood Pressure and Red Blood cell value fluctuation in non-HIV datasets affects
the classifier accuracy. Therefore, non-HIV accuracy is less than HIV data. Below graph indicate CNN+PCA gives almost 99%
accuracy for stage classification.

Fig. 10: Analysis graph of ML and DL methods

All rights reserved by www.grdjournals.com 59

 Chronic Kidney Disease Stage Prediction in HIV Infected Patient using Deep Learning
(GRDJE/ Volume 6 / Issue 5 / 007)

V. CONCLUSION
For companies to make various choices, the classification of chronic renal diseases in patients afflicted with HIV is highly useful.
First, delete the noisy data and then apply the PCA algorithm for the collection of attributes. Used CN N for CKD designation,
which will enhance the precision compared with chronic kidney disease stages after application of the attribute collection. With
this proposed CNN + PCA method, reliable classification of CKD data would be enhanced. After calculating and splitting the
information into five levels, eGFR is subsequently classified into a CKD formula. Early diagnosis and decision-making will also
be effectively carried out. The move to a proposed model would produce a 99 percent accurate real-time classification. It will allow
people to get early care and save lives. The x-ray image can be used for static data for the precision of the battery.

REFERENCES
[1] C. I. Bagnis, J. E. Heron, and D. M. Gracey, “Contemporary issues and new challenges in chronic kidney disease amongst people living with HIV,” AIDS
Res. Ther., vol. 17, no. 1, pp. 1–13, 2020, doi: 10.1186/s12981-020-00266-3.
[2] Y. Liu, C. Liu, J. Qin, L. Chen, C. Feng, and eB. Chen, “A machine learning methodology for diagnosing chronic kidney disease,” IEEE Access, vol. 8, pp.
20991–21002, 2020, doi: 10.1109/ACCESS.2019.2963053.
[3] A. S. Anwar and E. H. A. Rady , “Prediction of kidney disease stages using data mining algorithms,” Informatics Med. Unlocked, vol. 15, no. March, p.
100178, 2019, doi: 10.1016/j.imu.2019.100178.
[4] M. N. Amin, A. Al Imran, and F. T. Johora, “Classification of Chronic Kidney Disease using Logistic Regression, Feedforward Neural Network and Wide
Deep Learning,” 2018 Int. Conf. Innov. Eng. Technol. ICIET 2018, pp. 1–6, 2019, doi: 10.1109/CIET.2018.8660844.
[5] S. D. Sudarsan,N. Chetty, K. S. Vaisla, “Role of attributes selection in classification of Chronic Kidney Disease patients,” pp. 1–6, 2016, doi:
10.1109/cccs.2015.7374193.
[6] E. Perumal , P. Arulanthu , “Predicting the Chronic Kidney Disease using Various Classifiers,” 4th Int. Conf. Electr. Electron. Commun. Comput. Technol.
Optim. Tech. ICEECCOT 2019, pp. 70–75, 2019, doi: 10.1109/ICEECCOT46775.2019.9114653.
[7] K. Shankar, G. Devika, P. Manickam, M. Ilayaraja, “Optimal Feature Selection for Chronic Kidney Disease Classification using Deep Learning Classifier,”
2018 IEEE Int. Conf. Comput. Intell. Comput. Res. ICCIC 2018, pp. 1–5, 2018, doi: 10.1109/ICCIC.2018.8782340.
[8] R. Shinde, S. John, R. Jadhav, A. Maurya, R. Wable, and R. Dakshayani, “Chronic Kidney Disease Prediction and Recommendation of Suitable Diet Plan by
using Machine Learning,” 2019 Int. Conf. Nascent Technol. Eng. ICNTE 2019 - Proc., no. Icnte, pp. 1–4, 2019, doi: 10.1109/ICNTE44896.2019.8946029.
[9] S. C. Jat . R. Yadav, “Feature selection and dimensionality reduction methods for chronic disease prediction,” Int. J. Sci. Technol. Res., vol. 9, no. 4, pp.
2912–2918, 2020.
[10] S. saraswat, S. Vashisth, I. dhall, CKD diagnosis using Multilayer Perceptron Classifier’ 10th International Conference on Cloud Computing, Data Science
& Engineering (Confluence)” 2020.
[11] K. Shankar, M. Elhoseny, J. Uthayakumar, “Intelligent Diagnostic Prediction and Classification System for Chronic Kidney Disease,” Sci. Rep., vol. 9, no.
1, pp. 1–14, 2019, doi: 10.1038/s41598-019-46074-2.
[12] M. H. Abd Wahab, A. A. Johari, A. Mustapha, “Two-Class Classification: Comparative Experiments for Chronic Kidney Disease,” 2019 4th Int. Conf. Inf.
Syst. Comput. Networks, ISCON 2019, pp. 789–792, 2019, doi: 10.1109/ISCON47742.2019.9036306.
[13] G. Saini, B. Pandey, A. Khamparia, S. Tiwari, D. Gupta, and A. Khanna, “KDSAE: Chronic kidney disease classification with multimedia data learning
using deep stacked autoencoder network,” Multimed. Tools Appl., 2019, doi: 10.1007/s11042-019-07839-z.
[14] S. Maheshwari, S. Singhal, and M. Meena, Recent Findings in Intelligent Computing Techniques, vol. 707. Springer Singapore, 2019.
[15] V. Sharma, S. Sharma, A. Sharma, “Performance Based Evaluation of Various Machine Learning Classification Techniques for Chronic Kidney Disease
Diagnosis,” 2016, [Online]. Available: http://arxiv.org/abs/1606.09581.
[16] "Learn More About Kidney Disease: Types, Symptoms, and Causes, Siemens-healthineers.com, 2021. [Online]. Available: https://www.siemens-
healthineers.com/en-us/clinical-specialities/kidney-disease/about-kidney-disease. [Accessed: 04- Mar- 2021]
[17] Estimated Glomerular Filtration Rate (eGFR), National Kidney Foundation, 2021. [Online]. Available: https://www.kidney.org/atoz/content/gfr. [Accessed:
04- Mar- 2021]
[18] The dawn of a new era in chronic kidney disease, Astrazeneca.com, 2021. [Online]. Available: https://www.astrazeneca.com/what-science-can-do/stories/the-
dawn-of-a-new-era-in-chronic-kidney-disease.html. [Accessed: 04- Mar- 2021]
[19] 10 Applications of Deep Learning in Business - Algorithm-X Lab, Algorithm-X Lab, 2021. [Online]. Available: https://algorithmxlab.com/blog/10-
applications-deep-learning-in-business/. [Accessed: 04- Mar- 2021]
[20] Check out this amazing template to make your presentations look awesome at, Slideteam.net, 2021. [Online]. Available: https://www.slideteam.net/examples-
of-deep-learning-applications-ppt-powerpoint-presentation-introduction.html. [Accessed: 04- Mar- 2021]
[21] Wang, Huafeng et al. ‘A Hybrid CNN Feature Model for Pulmonary Nodule Malignancy Risk Differentiation’. 1 Jan. 2018 : 171 – 187.
[22] How Deep Learning Can Help Prevent Financial Fraud, Investopedia, 2021. [Online]. Available: https://www.investopedia.com/terms/d/deep-learning.asp.
[Accessed: 04- Mar- 2021]

All rights reserved by www.grdjournals.com 60