2023 IEEE IAS Global Conference on Emerging Technologies (GlobConET)

Loughborough University, London, United Kingdom. May 19-21, 2023

Chronic Kidney Disease Prediction Using Machine

Learning Algorithms and the Important Attributes
for the Detection
2023 IEEE IAS Global Conference on Emerging Technologies (GlobConET) | 979-8-3503-3179-0/23/$31.00 ©2023 IEEE | DOI: 10.1109/GlobConET56651.2023.10149900

Garima Shukla1 Gaurav Dhuriya2

Department of Electronics and Communication Engineering Computer Science Engineering
Noida Institute of Engineering and Technology, G. L Bajaj Institute of Technology and Management,
Greater Noida, India Greater Noida, India
[email protected] [email protected]

Sofia K Pillai3 Aradhna Saini4

Computer Science Engineering Computer Science Engineering
Noida Institute of Engineering and Technology, G. L Bajaj Institute of Technology and Management,
Greater Noida, India Greater Noida, India
[email protected] [email protected]

Abstract - In the modern world, chronic kidney disease has patient also cannot detect it. In the Fourth stage of the CKD,
become one of the most hazardous diseases. CKD is a condition the error is significantly increased and the patient starts to
in which the kidney cannot perform the proper filtering of the have some serious symptoms of kidney-related issues, and if
blood or it stooped working completely which causes the left the patient ignores this problem that leads to the Fifth stage
toxic into the blood, which leads the patient to death. It is likely of the CKD in which the kidney stop working completely.
impossible to detect CKD in the early stages, and it is very The only possible way to save the patient’s life in the fifth
difficult to save patient’s lives in the last stage of CKD. A stage is the renal transplant, which is very difficult as well as
patient's life can be saved by renal transplant or the early expensive. Hence, for saving a patient’s life from CKD is to
detection the CKD. Machine Learning algorithm techniques somehow to detect in its early stages. Where the nephrologist
have played a very important role in CKD prediction. Past can give the medicine to stop the infection in the kidney. We
medical test, reports can also be used as a tool for the early
can also use the past medical records for the detection of the
detection of renal disease. Machine Learning (ML) Techniques
like KNN, Decision Tree, and ANN are used in this review. We
CKD. The past medical record consists a lot of information
have to find out that Decision Tree has shown the best result of about the patient like age, blood pressure, sugar, hemoglobin,
98.60% of accuracy. Generally, the majority of the algorithms anemia and etc. these attributes can also be very useful for
are based on supervised learning and classification problem CKD detection because the human’s organs are
solving. We have explained some important attributes, which interconnected to each other and perform the task with help
play a major role in early CKD Prediction and Detection. of others. Like we can say that blood is pumped by the heart
Every attribute has its specific effect on CKD. The previous through all over the body and the kidney filters the blood and
researchers have done many experiments to get the best eliminates the toxic form it, hence if there is any problem
attribute and best ml technique for the prediction. in this with the heart the blood flow will be changed and that will
paper, we have studied all related ML techniques, and affect the kidney filtering. Smoking and depression can also
important Attributes and discussed the measurement factor for affect kidney filtering. Hence, we can use all the past records
CKD prediction. for the future detection of the CKD [3].
Keywords: CKD, Renal Disease, Machine Learning, Neural
network, GFR, Diabetes Mellitus, Cardio Vascular disease
(CVD), Creatinine.

I. INTRODUCTION
Chronic kidney disease is a life-threatening condition in
which the kidney is not able to filter the Blood or it is totally
failed to perform any task, then the toxic in the blood
remains in the blood, which kills the patient slowly, Inability
of any Prior test can detect the CKD in its early stages. The
only tests for CKD are the GFR test which is very expensive
but not able to detect the early stages of the CKD, and these
tests are not so common they take some specific test labs,
which are not available easily to very patients. The CKD is
Generally divided into the 5 stages based on the glomerular Fig. 1. Stages and infection of the CKD
filtration rate (GFR) [figure 1] [8]. In the First stage of CKD,
kidneys do some insignificant errors in blood filtering. In the II. LITERATURE REVIEW
second stage of the CKD, the error is increased a little bit
which the patient cannot detect. In the third stage of CKD, CKD is a one of the most sever disease, which is creating
the error in the blood filtering is more increased but the so many life losses. More than 10% of people are facing this

979-8-3503-3179-0/23/$31.00 ©2023 IEEE 1

Authorized licensed use limited to: Visvesvaraya Technological University Belagavi. Downloaded on November 22,2023 at 07:51:40 UTC from IEEE Xplore. Restrictions apply.
 critical renal problem. For the future prediction there are say that high aged patients have more chances of CKD
many researchers have done on the CKD using Machine infection [9]. (age distribution chart figure 2)
Learning Techniques [13], previously so many of the ML
algorithms were used for the prediction of CKD. ML
Techniques used for the CKD prediction and detection are
listed below:
A. K-Nearest Neighbor
It is known as the KNN technique. It is a classification
Algorithm and part of supervised learning. It is used for
knowledge Extraction such as classification and regression
problems. Many researchers for classification work use
KNN. Bashir et al. [1] created a framework using multi-layer
classification for the detection of diseases. After implying
KNN they got 57.41% of accuracy using a dataset from
startling and while using a dataset of Cleveland they got
58.42% of accuracy. Hasi et al. [2] have implied KNN on Fig. 2. Age Distribution Chart
Pima Indians dataset for CKD their research shown KNN has
generated 76.96% of accuracy with very less error rate for 2) Blood Pressure
Chronic Kidney Disease detection. The Pima Indians dataset BP is measured in f numerical values. The Measurement
has 768 samples, which include 8 numerical value attributes. unit of the Blood Pressure is millimeters of mercury (mmHg)
Khan et al. Used KNN in UCI Machine Learning Repository in our blood vessels. Blood Pressure is can be defined as the
for Lever infection detection, their result of KNN produced pressure that is generated by the Blood flow in or blood
an accuracy of 62.90% and the error rate of 0.3718. veins. In general, if BP is increased in a patient’s body that
can trigger the CKD or increase the chances of CKD
B. Decision Tree occurrence [10] [14]. (Blood Pressure distribution chart
A decision tree (DT) is a supervised learning method that figure 3).
may be applied as both a classification and regression tool.
Generally, it is structured as a tree classifier where internal
nodes consider as data set features, branches serve as
decision rules and each leaf node presents the result. Sharma
et al. [4] have used a decision tree for the Chronic Kidney
Disease prediction and got 98.60 % of accuracy, specificity
of 1, and precision rate of 1, by using the UCI Machine
Learning Repository dataset. Asker et al. [5] used Decision
Tree for renal disease prediction using 24 features from the
UCI ML Repository data set. They got 98.4127% of
accuracy and an error rate of 0.0159 [15].
C. Artificial Neural Network Fig. 3. Blood Pressure Distribution Chart
Artificial Neural Network is a sub-type of artificial
intelligence and works as a supervised learning method. 3) Specific Gravity
Artificial Neural Network has the neurons like a human Specific Gravity can be used for the measurement of the
brain, and every neuron is connected with other neurons in- kidney’s capacity to dilute or concentrate urine into plasma
network and these neurons are known as nodes. Artificial because urine is a solution of minerals, Salts, and compounds
Neural networks have 3 layers. The input layer is the top dissolved in water. The urine is more concentrated the higher
layer, the concealed layer is the next, and the output layer is the specific gravity. If the specific gravity is less it means
the last. The input layer retrieves input from the dataset that less concentrated urine, Hence kidneys need to work harder
is sent to Hidden Layer. Hidden Layer processed the input for the removal of the toxins. Then there are more chances
data and find the hidden pattern from this it is sent to the for renal damage or CKD [11]. (sg distribution chart figure
Output Layer where the result is generated. Chithra et al. [6] 4)
used the ANN for the CKD prediction over TCIA dataset
where they got 95.20% of accuracy and after using UCI ML
Repository data set they got 91.92% of accuracy while taking
full features in the dataset. Vijayrani et al. [7] used ANN for
Renal disease prediction. They collected the data from
different labs, hospitals, and medical centers. They achieved
87.70% accuracy in their experiments.
1) Age
The age is a numerical value, which is measured in the
form of the year. In previous studies we have found that
people of age group 45+ have more chances of suffering
from renal Disease. Age is a very important attribute in the Fig. 4. Specific Gravity Distribution Chart
case of CKD [12]. So on the basis of early studies, we can

2
Authorized licensed use limited to: Visvesvaraya Technological University Belagavi. Downloaded on November 22,2023 at 07:51:40 UTC from IEEE Xplore. Restrictions apply.
 III. DISCUSSION TP = number of true positives
We have so many of ML and AI techniques, which are TN = number of true negatives
used for the early detection, and prediction of the renal FP = number of false positives
disease. In addition, the past records can help to get the FN = number of false negatives
diseased outcome. Hence, we have some Measurement E. Recall
method for calculating the results we got after performing the
techniques on the attributes. The Measurement Methods are It is known as the average probability of total retrieval,
listed below- and calculated by-

A. Mean Absolute Error: 

‡ ƒŽŽ ൌ 
MAE is an amount to calculate the prediction, which  ൅ 
deviate from the true probability. P(j,k) is the approximate of F. Precision
coefficient to be class k. MAE takes value between 0 to 1, It is known as related number between the accurately
and is given by- classified module to whole classified error-prone modules.
σୟୀଵ ȁ›ୟ  െ  šୟ ȁ Precision is the number of accurately prescribed as error:
 ൌ
 
MAE = Mean absolute error ”‡ ‹•‹‘ ൌ
 ൅ 
›ୟ  = Prediction
G. F-Measure
šୟ = True value
N = total number of data points It is one of union of both Recall and Precision. F-
Measure uses the retrieval information to approximate
B. Root Mean Squared Error classification performance and is given by:
RMSE is the square root of the mean span between ୘୔
F- Measure = భ
predicted and remarked data. Signifying the approximated ୘୔ାమሺ୊୔ା୊୒ሻ
probability of the acceptable data j and the selected value for
the acceptable data k by p(j,k) and Tk independently, RMSE H. Accuracy
is calculated as follow- It is given by the percentage of accurately classified
cases, as well as, accuracy is one of the most useful
σౖ ෞഠ ିୟ౟
೔సబ ୟ
RMSE = ට classification matrix in classification data and calculated by-
୞
 ൅ 
 —”ƒ › ൌ
RMSE = root mean square error  ൅  ൅  ൅ 
I = variable I Type of relation and inter-relation between themselves
and CKD [figure 2]. An age distribution Chart [figure 2] is
ƒෝ=
న estimated time series also created by the tableau for the better understanding of the
ai= actual observations time series dataset and age effect in the CKD. We gave generated the
Z= number of non missing data points distribution chart for the every single attribute in figure 31.
C. Relative Absolute Error
RAE is the way to help in prediction of performance of
different models. This is used in different area like machine
learning, data mining, etc. Using below formula, it can be
calculated-
ୟ  െ  ୣ
Ɂ ൌ  ฬ ฬ
ୣ

Ɂ = percent error

ୟ  = actual value note
ୠ = expected value
D. Root Relative Squared Error (RRSE)
RRSE is used for basic level predicted. It takes total
squared error

ଶ
σ୬୨ୀଵ൫ሺ୧୨ሻ  െ  ୨ ൯
୧ୀ ඩ
ഥ ൯ଶ
σ୬୨ୀଵ൫୎  െ  
ሺ୧୨ሻ = Predicted value by model
୨ = Target value
୬
ͳ
ഥ  ൌ  ෍ ୨

 Fig. 5. Data Flow Diagram
୨ୀଵ

3
Authorized licensed use limited to: Visvesvaraya Technological University Belagavi. Downloaded on November 22,2023 at 07:51:40 UTC from IEEE Xplore. Restrictions apply.
 TABLE I. ACCURACY TABLE very few test for the CKD detection, which are not so much
S No. ML Techniques Accuracy(%) available in every region and are very much expensive too.
1 KNN 76.96 In the last stage of the CKD renal transplant in the only way
2 Decision Tree 98.60 save patient life and the surgery success rate is very low.
3 ANN 95.20 Hence, the early detection of the renal ways is the best cure
for this life-threatening problem. ML and techniques are the
best possible solution for this problem, which is also
developing day by day. We can also use the past medical
reports for the CKD detection, they include many important
attributes which are used by the nephrologist and researchers.
Knowledge of these attribute can give a lot of idea about the
kidney condition and could be used as the early detection
method for the CKD.
REFERENCES
[1] Bashir, S., Qamar, U., Khan, F. H., & Naseem, L. (2016). HMV: a
medical decision support framework using multi-layer classifiers for
disease prediction. Journal of Computational Science, 13, 10-25.
[2] Hashi, E. K., Zaman, M. S. U., & Hasan, M. R. (2017, February). An
expert clinical decision support system to predict disease using
classification techniques. In 2017 International Conference on
Electrical, Computer and Communication Engineering (ECCE) (pp.
Fig. 6. Accuracy Graph 396-400). IEEE.
[3] Khan, B., Naseem, R., Ali, M., Arshad, M., & Jan, N. (2019).
TABLE II. MOST IMPORTANT ATTRIBUTES Machine learning approaches for liver disease
diagnosing. International Journal of Data Science and Advanced
S.N Attribut Descriptio Class Measure Scale Analytics (ISSN 2563-4429), 1(1), 27-31.
o e n [4] Sharma, S., Sharma, V., & Sharma, A. (2016). Performance based
1 sg Specif Nomina 1.005,1.010,1.015,1.020,1.0 evaluation of various machine learning classification techniques for
Gravity l 25 chronic kidney disease diagnosis. arXiv preprint arXiv:1606.09581.
[5] Alasker, H., Alharkan, S., Alharkan, W., Zaki, A., & Riza, L. S.
TABLE III. AVERAGE IMPORTANT ATTRIBUTE (2017, October). Detection of kidney disease using various intelligent
classifiers. In 2017 3rd international conference on science in
S.No Attribute Description Class Measure scale information technology (ICSITech) (pp. 681-684). IEEE.
1 bp Blood Numerical In mm/Hg [6] Chittora, P., Chaurasia, S., Chakrabarti, P., Kumawat, G.,
Pressure Chakrabarti, T., Leonowicz, Z., & Bolshev, V. (2021). Prediction of
Chronic Kidney Disease-A Machine Learning Perspective. IEEE
TABLE IV. LEAST IMPORTANT ATTRIBUTE Access, 9, 17312-17334.
[7] Vijayarani, S., Dhayanand, S., & Phil, M. (2015). Kidney disease
S.No. Attribute Description Class Measure prediction using SVM and ANN algorithms. International Journal of
Scale Computing and Business Research (IJCBR), 6(2), 1-12.
1 age Age in year Numerical Age in [8] Khan, B., Naseem, R., Muhammad, F., Abbas, G., & Kim, S. (2020).
An empirical evaluation of machine learning techniques for chronic
Year kidney disease prophecy. IEEE Access, 8, 55012-55022.
[9] Prakash, S., & O'Hare, A. M. (2009, September). Interaction of aging
and chronic kidney disease. In Seminars in nephrology (Vol. 29, No.
5, pp. 497-503). WB Saunders.
[10] Constantiner, M., Sehgal, A. R., Humbert, L., Constantiner, D., Arce,
L., Sedor, J. R., & Schelling, J. R. (2005). A dipstick protein and
specific gravity algorithm accurately predicts pathological
proteinuria. American journal of kidney diseases, 45(5), 833-841.
[11] Menon, V., Wang, X., Greene, T., Beck, G. J., Kusek, J. W.,
Marcovina, S. M., & Sarnak, M. J. (2003). Relationship between C-
reactive protein, albumin, and cardiovascular disease in patients with
chronic kidney disease. American Journal of Kidney Diseases, 42(1),
44-52.
[12] Navaneethan, S. D., Schold, J. D., Jolly, S. E., Arrigain, S.,
Winkelmayer, W. C., & Nally Jr, J. V. (2017). Diabetes control and
the risks of ESRD and mortality in patients with CKD. American
Journal of Kidney Diseases, 70(2), 191-198.
[13] Malhotra, R., Craven, T., Ambrosius, W. T., Killeen, A. A., Haley,
W. E., Cheung, A. K., & SPRINT Research Group. (2019). Effects of
intensive blood pressure lowering on kidney tubule injury in CKD: a
longitudinal subgroup analysis in SPRINT. American Journal of
Kidney Diseases, 73(1), 21-30.
[14] Umanath, K., & Lewis, J. B. (2018). Update on diabetic nephropathy:
Fig. 7. Relation between bp and age core curriculum 2018. American Journal of Kidney Diseases, 71(6),
884-895.
[15] Sobrinho, A., Queiroz, A. C. D. S., Da Silva, L. D., Costa, E. D. B.,
IV. CONCLUSION Pinheiro, M. E., & Perkusich, A. (2020). Computer-aided diagnosis of
Recent study have shown that CKD is one of the most chronic kidney disease in developing countries: A comparative
analysis of machine learning techniques. IEEE Access, 8, 25407-
sever disease. Inability of the early detection tests are the
25419.
reason for the high death rate in CKD. Nephrologist have

4
Authorized licensed use limited to: Visvesvaraya Technological University Belagavi. Downloaded on November 22,2023 at 07:51:40 UTC from IEEE Xplore. Restrictions apply.