Early Warning Systems for Chronic Kidney Disease:

A Data-Driven Approach Using Machine Learning

Rohith Indupalli Laahiri Gurram Sudharsan Uppala

Department of CSE Department of CSE Department of CSE
Vignan’s Foundation for Science, Technology Vignan’s Foundation for Science, Technology Vignan’s Foundation for Science, Technology
and Research and Research and Research
Guntur, India Guntur, India Guntur, India
rohith.indupalli@gmail.com gurramlaahiri629@gmail.com sudharsanuppala2005@gmail.com

Nagi Reddy Annapureddy Dr. Deva Kumar S line 1: 6th Given Name Surname
Department of CSE Associate Professor line 2: dept. name of organization
Vignan’s Foundation for Science, Technology Department of CSE (of Affiliation)
and Research Vignan’s Foundation for line 3: name of organization
Guntur, India Science, Technology and (of Affiliation)
naagireddy@gmail.com Research line 4: City, Country
Guntur, India line 5: email address or ORCID
drsdk@gmail.com

Abstract—Acute renal health, frequently referred to as acute

kidney wellness, is a state that positively affects both the kidneys effectiveness of medical tests, medications, and surgical
and the individual. If left untreated, this condition can develop treatments as well as to find patterns in vast amounts of
into end-stage renal failure, which can be fatal. Machine learning clinical and diagnostic data[2].
techniques have emerged as vital instruments in medicine and Although there isn’t a single underlying reason for chronic
are critical for illness prediction. The prognosis of chronic
kidney disease (CKD), the decline is frequently permanent
renal illnesses requires the formation and verification of a
forecast model. 400 chronic renal disease patient samples were and causes major Wellness solutions. Over the last ten years,
chosen Extracted from the UCI Machine Learning Repository, the Kidney Disease Outcomes Quality Initiative by the US
this study employs four machine learning classifiers: Support National Kidney Foundation Glomerular filtration rate (GFR)
Vector Machine (SVM), Multilayer Perceptron (MLP), K- below 60 mL/min/1.73 m² has developed The initial standard
Nearest Neighbor (KNN), and Naive Bayes (NB).model’s
that characterized chronic kidney disease (CKD), irrespective
performance is improved by applying the bagging ensemble
approach. The selected machine The learning The classifiers are of its origin, by the existence of pathological abnormalities or
then trained utilizing the combined data from the chronic renal imaging abnormalities for three more months [3]. Businesses
disease dataset. The nonlinear features and classifications are attempt to support new business strategies based on the
then used to create the Kidney Disease Collection. classification scores in order to better serve their customers.
The decision-making process in opinions of the general
Keywords—chronic renal disease, classification algorithms,
Ensemble learning, Dimensionality reduction people regarding present government initiatives[4].
Choosing the optimal driver topology to boost the device’s
performance was the aim of the endeavor[5]. Clinical
I. INTRODUCTION information and laboratory testing are usually the first steps
in the diagnosis of CDK. Imaging tests, followed by a biopsy.
The emerging area of computational health informatics Despite being the most used diagnostic procedure since
encompasses a wide range of sciences, Encompassing then, biopsy has some drawbacks, including being invasive,
biomedical, healthcare, nursing, IT, computer science, costly, time-consuming, and even dangerous.For instance, the
and statistics. It employs data mining methods to find patient may experience an infection, anxiety over surgery,
patterns in vast amounts of clinical and diagnostic data or an incorrect diagnosis during the biopsy[6]. Clinical
to forecast the efficacy of medical tests, medications, and information and laboratory testing are usually the first steps
surgical operations[1]. In the new field of computational in the diagnosis of CDK.Imaging procedures, followed by
health informatics research, the disciplines of ”Biomedical, a biopsy. Despite being the most used diagnostic procedure
healthcare, nursing, IT, computer science, and statistics are since then, biopsy has some drawbacks, including being
all involved. It Utilizes data mining methods to predict the invasive, costly, time-consuming, and even dangerous. For
instance, the patient may experience an infection, anxiety
over surgery, or an incorrect diagnosis during the biopsy[7].
Chronic Renal Failure, another name for Chronic Kidney With a success rate of over 93%, [6] examined the use of
Disease, is a global concern today. Persistent renal illness the Support Vector Machine technique and health indicators
entails the steady decrease in kidney performance over many for acute kidney health individuals. findings on the testing
years. A condition known as Persistent renal illness arises when set’s metrics, sensitivity, and specificity. The multilayer
the kidneys are damaged, and the body fails to filter efficiently perceptron model was used by [7] to create a decision support
out toxins. When kidney function falls below 25% of system for renal illness diagnosis.
normal, a person is diagnosed with Chronic renal illness, In order to uncover passed time, the authors [8] analyzed a
recognized as one of the significant global health case of a single CKD patient using dimensionality reduction
concerns[8].Due to a lack of unbounded sensibility, CKD is algorithms like ICA and PCA. [9] 2020 He proved that Naive
caused by fewer sensations, which most people in rural regions Bayes categorization was superior. One of the crucial
are unaware of.Although technology is developing quickly, characteristics of the Bayes theorem is that naive Bayes
individuals are not paying enough attention.They run a serious classifiers are probabilistic classifiers. Random Forest
danger of kidney damage in this way[9]. outperformed Naive Bayes in our study. In addition, Naive
This research offers A technique for evaluating the risk Bayes’ delicacy rate was lower than that of other styles, which
factors for chronic kidney disease (CKD) and suggests was 93.9056.
warning patients about them so they can take care of This relates to the accuracy of the KNN in this system.A
themselves. In general, this study may make it easier for classifier performs better than one that uses a decision tree
the doctor to identify the symptoms and treat them early classifier. The procedure that the author suggested would
on. The risk factors will be predicted using A variety of automatically. Assess and calculate the effects of a patient’s
algorithms, such as Na¨ıve Bayes, Support Vector Machine, kidney disease [10].
and K-Nearest. Neighbours, and MLP. Using the CKD- 15 datasets, an intelligent opinion sys-
II. LITERATURE REVIEW tem erected on an ensemble approach achieved 96 delicacy,
according to inside the machine learning order [11]. A
Many researchers use KNN extensively for classification dataset of CKD with 400 cases and 25 characteristics was
challenges. Using the same dataset as their reference Pima used by [12]. The three main indicators of whether or not a
Indians data, [1] utilized KNN for CKD. The findings show person has an illness are haemoglobin, albumin, and specific
that KNN predicts CKD with 76.96% accuracy and a minimal gravity. As a result, these were discovered, included in the
error rate. [2] used A dataset from the UCI ML repository to data set, and selected using filtering features that could
forecast liver illness using KNN. Their KNN results show an achieve a high degree of accuracy. They trained their data
accuracy of 62.90%. The accuracy and error rates were set after feature selection, and he performed cross-validation,
0.3718, as was previously mentioned. achieving 99.1% accuracy. In order to categorize the
With neural networks included to complete the absent val- issue, The Support Vector Machine (SVM) technique is
ues, the combination of multilayer perception and preprocess- the most employed approach for generating predictions in
ing of the data set achieved a Precision of 0.995. in predicting data mining for this type of data. The primary aim of
the early stages of CKD. The technique consists of removing the Support Vector Machine (SVM) is to determine the
outliers, selecting the top seven qualities, which includes best hyperplane that separates two categories worth of
statistics, and removing the attributes that have the highest training data [13]. The classification technique known as
correlation, which is achieved by using principal component K-nearest neighbors (KNN) Seeks the nearest data points
analysis (PCA) [3]. NB has been developed by [1] for renal in the feature space to categorize unfamiliar samples.[14].
The UCIML repository provides the data set. There are 400 Support vector machines (SVMs) are a common data
people in the population with 24 traits in this data collection. mining technique for classifying data and predicting its
There are 250 people in the early stages of CKD and 150 category . Developing a hyperplane in the training data that
people without CKD in the population. They achieved maximally separates the two classes is the main goal of
99.3671% accuracy with an error rate of 0.0057 using all 24 SVM [15].
features. When building an ideal ensemble machine, an astounding
NB was utilized by [4] to author pro thesis for gender and 96.5% accuracy rate has been demonstrated. Predictive
age using several feature types. They came to the conclusion learning model for chronic renal disease. combination of
that the accuracy of NB employing Sn-gram (POS) and algorithmic slimming and random forest[16].Riddhi suggested
gender Tn-gram (POS) characteristics with hotel review a kidney disease diagnosis system based on machine learning.
prediction is 60.6%. The accuracy of NB for each feature is Adaboost, gradient boost, logistic regression, bernoulli
50.4%. SVM was used by [5] to classify liver patients using naive bayes, and random forest have all been widely used
data from the UCI ML repository. They obtained 71.5026% algorithms. Dealing with the null values based on appropriate
accuracy over the same original dataset, but 68.64% accuracy imputation was a difficult task because the dataset included a
after oversampling. relatively low number of data records and several null values,
necessitating correct pre-processing[17].
Through the application of cutting-edge methods, this study
seeks to change the initial identification and progression sources of the Data utilized in this research.
tracking of chronic renal disease. Recursive Feature Prepare the information: The majority in clinical data will
Elimination (RFE) is applied alongside the ensemble contain inaccurate or missing values. Either the mean or the
approach to pinpoint essential attributes required for a median is used to fill in the missing numbers.
successful diagnosis of chronic kidney Illness. The objective Data Transformation: Blood pressure, age, and numerous
of the research endeavor is to employ the XGBoost algorithm other health metrics will be standardized or normalized. These
and k-fold cross-validation to forecast the trajectory. CKD are some of the changes that must be made in order to train
with unparalleled precision using tightly controlled real- the model with all of the features considered equally.
time data sets. As a result, it was believed that this study Feature Selection: Recursive feature elimination, PCA, or
would significantly advance CKD tracking and time-to-event feature selection techniques would be used to minimize the
prediction, ensuring better patient outcomes and increased dimension of the data. Prior research shows that applying
medical sector efficiency. A basis for future predictive PCA enhances the model’s ability to classify CKD patients.
analytics research has been established by this study[18].
To prevent renal failure progressing to chronic kidney disease,
early prediction is crucial for both patients and professionals.
The suggested frameworks for this study were assembled
employing two feature selection techniques (RFECV and
UFS) and three machine-learning algorithms (RF, SV, and
DT). Tenfold cross-validation was implemented for evaluating
the frameworks. TheInitial datasets encompassing all 19
features were analyzed using the four machine learning
techniques: RF, SVM, and XGBoost yielded the highest
accuracy when we used the models on the original dataset.
”The precision was 82.56% for the quintuple-class and 99.8%
for the binary. Performance-wise, RF fared better than DT.
The highest f1 score values were also generated via RF[19].
Random forest classifiers will lower the amount of features
in the prediction algorithm, and it may be possible that some
medical tests are not necessary in some situations where other
variables must be used to Complete the missing information.
It can be acquired by completing the missing data and
applying the other variables. This innovative approach
predicts the CKD status based on attributes and includes a
number of components, such as feature selection, data
preparation, and handling missing information. Two
outstanding algorithms that rely solely on the features
employed are random forests and decision trees. Both
methods have a high accuracy value. This study demonstrates
that domain expertise is essential for deciphering CKD
Figure 1 : Proposed Architecture
clinical data[20].

2. Model Selection
III. METHODOLOGY
Predicting Chronic Kidney Disease (CKD) is the primary
A CKD prediction system Based on the implementation of focus of the model-building stage, which comes after
the machine learning model proposed in this research. system preprocessing the data. Once missing values have been
is intended to use a range of machine learning methods, addressed, categorical variables have been encoded, and
including KNN, SVM, NB, and MLP, to concentrate on the numerical variables have been scaled, all of these procedures
early diagnosis of CKD. To enhance precision, sensitivity, will be utilised to create models that use machine learning
specificity, an ensemble approach will also be developed to make predictions. Typical models include SVM, gradient
to aggregate predictions from various models. After that, it boosting method, logistic regression, decision trees, and
preprocesses the patient data so that its MLPs, including random forests. A portion of the data will be used to train
those with CKD status, can be trained as shown in Figure 1. each model in order to identify the links and patterns present
in the features.
1. Data Gathering and Preparation
Data Source: Public clinical databases, the UCI CKD a. KNN, or K-Nearest Neighbours
dataset, or hospital databases that hold patient records in- Data points are categorized by the non-parametric
cluding demographics, test results, and diagnoses will be the KNN algorithm according to how close they are to other
points in
the feature space. The model will be educated on a subset 4. Model Assessment
of the data collection, and its performance will be reviewed Performance Metrics: The effectiveness of all the
using measurements such as accuracy, precision, recall, and models will be evaluated using accuracy, precision, recall, F1-
the F1 score. The optimal number of neighbours (k) to ensure score, and the area under the ROC-AUC curve. These metrics
accurate predictions will be found using cross-validation. are selected because they not only provide only measure the
actual success of the model for identifying CKD patients but
b. Support vector machines, or SVMs also rule out false positives or negatives.
SVM is a important bracket fashion that Accuracy: it tests the overall efficiency of the model in
chooses the most effective hyperplane for class division. The yielding highly accurate prediction.
model’s performance will be estimated grounded on its Accuracy and Recall: A good performance measure that
capability to generalise to new data. To find the stylish speaks to a balance between true positives and avoiding
configuration for the CKD dataset, a range of kernel functions false negatives, particularly of great importance in the CKD
will be assessed with a focus on delicacy and other bracket detection.
criteria . ROC-AUC: It is the trade-off between true positives and false
positives, hence at multiple levels of thresholds.
c. Naive Bayes( NB) Comparison of Models: Comparing the models according
Founded on Bayes’ theorem, the Naive Bayes to the performance metrics; because in this case the taken
classifier assumes that the predictors are independent of one benchmark is NB and therefore the outcome in more complex
another. The model will be trained on the dataset, and its models, with a combination of SVM along with KNN & NB,
predictive performance will be evaluated using Bayes theorem with its previous studies, prove yielding higher accuracy than
criteria. Given the simplicity and effectiveness of Naive every of these models.
Bayes, it serves as a strong birth for comparison against more
complex models. 5. Ensemble Methods
To increase the precision of prediction accuracy levels, we
d. Perceptron with Multiple Layers (MLP) shall employ ensemble methods. A hybrid model of KNN,
One kind of artificial neural network that works well SVM, and NB will be used. For such models, Majority
for simulating intricate relationships in data is MLP. It is Voting or Stacking techniques for aggregating predictions can
made up of several interwoven layers of neurones.The model be performed.
will be trained and evaluated using standard metrics, with an
emphasis on performance optimisation through the
manipulation of hyperparameters like learning rate and hidden IV. RESULTS AND ANALYSIS
layer count. Several machine learning models, including SVM, NB,
KNN, and MLP, were trained during the experimentation
3. Dimensionality Reduction stage. A number of Performance indicators are employed to
Dimensionality reduction is the lowering of high- substantiate the findings for these models. To determine
dimensional data space by representation, with more whether any model can accurately predict CKD, performance
emphasis on the most informative features that help to: is evaluated using the degree of accuracy, precision, recall,
 Accelerate computations and F1 score. The results indicate that while each model has
 Lower the chances of its advantages, the ensemble model outperforms the others
overfitting Enhance due to its utilization of SVM, NB, KNN, and MLP
visualization predictions. The ensemble model’s success stems from its
 Remove noise capacity to leverage the capabilities of various classifiers,
 Make the data easier to interpret which enhances robustness against overfitting and enables the
capture of more intricate patterns in the dataset.
Following individual model evaluation, PCA will be used This structure was utilized by confusion matrices to show
to lower the dataset’s dimensionality. Principal components the performance of each model in terms of false positives,
are a new collection of features that PCA creates from the false negatives, true positives, and true negatives. The
original feature space in order to best capture the maximum graphical representation was essential for illustrating incorrect
variance in the data. Therefore, this phase is similar to classifications and demonstrated a distinct distinction between
lowering the curse of dimensionality, which speeds up the CKD and non-CKD patients, both of which needed to be
model’s training and improves performance overall. The improved.
training data will undergo PCA, and the test data will undergo
the same transformation. At least a sizable portion of the data We compared four classifiers—Support Vector Classifier
variance—let’s say 95%—will be preserved by the number of (SVC), Gaussian Naive Bayes (GaussianNB), K-Nearest
components kept. The models discussed above will then be Neighbours (KNN), and Multi-Layer Perceptron (MLP)—
run once more to observe the impact of dimensionality before and after using Principal Component Analysis (PCA)
reduction on their prediction capabilities. in order to assess the effect of dimensionality reduction on
model-performance.
 Table 1:Performance Metrics
Table 2:Hybrid Model Performance
Model Accuracy Precision Recall F1 Score
Hybrid Model
Na¨ıve Bayes 0.95 0.96 0.95 0.95
(SVM , KNN , NB)
Support Vector Ma- 0.85 0.85 0.85 0.85
chine Metric Value
MLP 0.62 0.39 0.62 0.48 Precision 0.98
KNN 0.93 0.93 0.93 0.92
Recall 0.98
F1-Score 0.98
Accuracy 0.98
The accuracy measurements showed that PCA's dimensionality
reduction had variable effects on model performance as showed
in Table 1. Notably, after PCA, the accuracies of the SVC and
MLP classifiers increased from 85% and 62% to 88% and 89%,
V. CONCLUSION
respectively, showing a slight improvement. On the other hand,
GaussianNB and KNN continued to perform similarly, In this study, we will explore the capabilities of various models in
demonstrating the resilience of their decision boundaries in both predicting chronic kidney disease. We will begin by analyzing each
high-dimensional and low-dimensional spaces. This model separately, ranging from SVM to Naive Bayes, KNN, and
investigation demonstrates how PCA can improve classifier MLP, using accuracy, precision, recall, and F1 score metrics to
efficiency, especially for models that stand to gain from smaller evaluate their predictive effectiveness for CKD. The application of
feature spaces by reducing computing complexity and PCA helped to decrease the dimensionality for a dimensionality
overfitting. reduction process in the subsequent step. PCA minimized the number
of features while preserving the essential information for more
efficient preprocessing. This preprocessing step is expected to
enhance model performance by reducing overfitting and
improving generalization to new data.

The concluding part of our analysis focused on model en-

sembling, where the outputs from various models were merged to
create an even more robust and precise prediction system. The
ensemble model outperformed all individual models that were
evaluated. Such outcomes demonstrate the advantage of ensemble
methods, as they leverage the strengths of different models to
enhance prediction accuracy. In summary, our findings suggest that
applying ensemble techniques in conjunction with suitable
preprocessing, like PCA, significantly enhances the accuracy of
CKD predictions, potentially aiding healthcare professionals in
early disease detection and timely treatment. Future efforts will
aim at further refining the ensemble strategy and exploring feature
Figure 2:Comparison of Metrics
mining.
Models for Ensemble learning: REFERENCES
a. KNN [1] E. K. Hashi, M. S. Uz Zaman, and M. R. Hasan, “An expert clinical de-
cision support system to predict disease using classification techniques,”
b. Naïve Bayes ECCE 2017 - Int. Conf. Electr. Comput. Commun. Eng., pp. 396–400, 2017,
c. Support Vector Machine doi: 10.1109/ECACE.2017.7912937.

[2] B. Khan, R. Naseem, M. Ali, M. Arshad, and N. Jan, “Machine Learning

Because of their higher accuracy for the dataset being Approaches for Liver Disease Diagnosing,” Int. J. Data Sci. Adv. Anal., vol.
studied, Naive Bayes (NB), Support Vector Machine (SVM), 1, no. 1, pp. 27–31, 2019, doi: 10.69511/ijdsaa.v1i1.71.
and K-Nearest Neighbors (KNN) were chosen for the hybrid
model. Each classifier has its own advantages: KNN enables [3] A. J. Aljaaf et al., “Early Prediction of Chronic Kidney Disease Using
efficient non-parametric learning based on neighborhood Machine Learning Supported by Predictive Analytics,” 2018 IEEE Congr.
closeness, SVM is excellent at establishing intricate decision Evol. Comput. CEC 2018 - Proc., pp. 1–9, 2018, doi:
10.1109/CEC.2018.8477876.
boundaries, and Naive Bayes delivers quick probabilistic
predictions as showed in Table 2. By combining these models [4] I. Ameer, G. Sidorov, and R. M. A. Nawab, “Author profiling for age and
into a hybrid framework, performance and generalization on gender using combinations of features of various types,” J. Intell. Fuzzy Syst.,
vol. 36, no. 5, pp. 4833–4843, 2019, doi: 10.3233/JIFS-179031.
unseen data are improved by utilizing their complementary
strengths to increase overall prediction accuracy and
robustness through ensemble techniques like weighted or [5] B. P. Singh, S. K. Goyal, S. A. Siddiqui, and P. Kumar, Intelligent
majority voting. Computing Techniques for Smart Energy Systems, vol. 607. 2020.
 [6] Y. Amirgaliyev, S. Shamiluulu, and A. Serek, “Analysis of Chronic Retinopathy,” Comput. Methods Biomech. Biomed. Eng. Imaging
Kidney Disease Dataset by Applying Machine Learning Methods,” Vis., vol. 11, no. 3, pp. 530–541, 2023, doi:
IEEE 12th Int. Conf. Appl. Inf. Commun. Technol. AICT 2018 - Proc., 10.1080/21681163.2022.2083020.
pp. 1–4, 2018, doi: 10.1109/ICAICT.2018.8747140.
[22] S. Deva Kumar, S. Venkatramaphanikumar, and K. Venkata Krishna
Kishore, “HIUNET: A Hybrid Inception U-Net for Diagnosis of Di-
[7] H. Yan, Y. Jiang, J. Zheng, C. Peng, and Q. Li, “A multilayer abetic Retinopathy,” Intell. Autom. Soft Comput., vol. 37, no. 1, pp.
perceptron-based medical decision support system for heart disease 1013–1032,2023,doi:10.32604/iasc.2023.0
diagnosis,” Expert Syst. Appl., vol. 30, no. 2, pp. 272–281, 2006, doi:
10.1016/j.eswa.2005.07.022.

[8] G. Kaur and M. Tech, “Mining Algorithms In Hadoop,” Proc. Int. Conf.
Inven. Comput. Informatics (ICICI 2017), no. Icici, 2017.

[9] M. A. Islam, S. Akter, M. S. Hossen, S. A. Keya, S. A. Tisha, and S.

Hos- sain, “Risk factor prediction of chronic kidney disease based on
machine learning algorithms,” Proc. 3rd Int. Conf. Intell. Sustain. Syst.
ICISS 2020, pp. 952–957, 2020, doi:
10.1109/ICISS49785.2020.9315878.

[10] W. H. S. Gunarathne, K. D. Perera, and K. K.A.D.C.P, “Performance

Evaluation on Machine Learning Classification Techniques for Disease
( CKD ),” pp. 291–296, 2017, doi: 10.1109/BIBE.2017.00056.

[11] N. I. Md. Ashafuddula, B. Islam, and R. Islam, “An Intelligent Diagnos-

tic System to Analyze Early-Stage Chronic Kidney Disease for Clinical
Application,” Appl. Comput. Intell. Soft Comput., vol. 2023, 2023, doi:
10.1155/2023/3140270.

[12] M. Almasoud and T. E. Ward, “Detection of chronic kidney disease

using machine learning algorithms with least number of predictors,”
Int. J. Adv. Comput. Sci. Appl., vol. 10, no. 8, pp. 89–96, 2019, doi:
10.14569/ijacsa.2019.0100813.

[13] R. G. Brereton and G. R. Lloyd, “Support Vector Machines for classi-

fication and regression,” Analyst, vol. 135, no. 2, pp. 230–267, 2010,
doi: 10.1039/b918972f.

[14] A. Charleonnan, T. Fufaung, T. Niyomwong, W. Chokchueypattanakit,

S. Suwannawach, and N. Ninchawee, “Predictive analytics for chronic
kidney disease using machine learning techniques,” 2016 Manag. Innov.
Technol. Int. Conf. MITiCON 2016, pp. MIT80–MIT83, 2017, doi:
10.1109/MITICON.2016.8025242.

[15] M. A. Estudillo-Valderrama et al., “A distributed approach to alarm

man- agement in chronic kidney disease,” IEEE J. Biomed. Heal.
Informatics, vol. 18, no. 6, pp. 1796–1803, 2014, doi:
10.1109/JBHI.2014.2333880.

[16] C. Choudhary, L. S. Nagra, P. Das, J. Singh, and S. S. Jamwal,

“Optimized Ensemble Machine Learning Model for Chronic
Kidney Disease Prediction,” Proc. - 4th IEEE 2023 Int. Conf.
Comput. Commun. Intell. Syst. ICCCIS 2023, pp. 292–297,
2023, doi: 10.1109/ICC- CIS60361.2023.10425073.

[17] R. Bhowmick and V. M. Arul Xavier, “Machine Learning Models

for Analysis and Prediction of Chronic Kidney Disease,” 2023
9th Int. Conf. Adv. Comput. Commun. Syst. ICACCS 2023, vol.
1, pp. 937–942, 2023, doi:
10.1109/ICACCS57279.2023.10112872.

[18] N. Sonone and A. Daniel, “Early Prediction and Progrssion of

Chronic Kidney Disease Using Machine Lerning Techniques,”
Proc. 2nd IEEE Int. Conf. Netw. Commun. 2024, ICNWC 2024,
pp. 1–6, 2024, doi: 10.1109/ICNWC60771.2024.10537571.

[19] D. A. Debal and T. M. Sitote, “Chronic kidney disease prediction

using machine learning techniques,” J. Big Data, vol. 9, no. 1,
2022, doi: 10.1186/s40537-022-00657-5.

[20] C. Kaur, M. S. Kumar, A. Anjum, M. B. Binda, M. R. Mallu, and

M. S. Al Ansari, “Chronic Kidney Disease Prediction Using
Machine Learning,” J. Adv. Inf. Technol., vol. 14, no. 2, pp. 384–
391, 2023, doi: 10.12720/jait.14.2.384-391.

[21] D. K. Salluri, V. Sistla, and V. K. K. Kolli, “HRUNET: Hybrid

Residual U - Net for automatic severity prediction of Diabetic