PREDICTION OF CHRONIC KIDNEY DISEASE USING MACHINE

LEARNING TECHNIQUES

Abstract - Early diagnosis and characterization are SVM, Gradient Boosting, Xgboost, Adaboost, Naive
essential in determining treatment for chronic kidney Bayes, Bagged Decision Trees, Voting Classifier,
disease (CKD). CKD is an ailment that damages the UCI Repository.
kidneys and affects the effective functioning of
I. INTRODUCTION
excreting waste and balancing body fluids. Some of
the complications included are hypertension, anemia Chronic kidney disease, or CKD, is a condition in
(low blood count), mineral bone disorder, poor which the kidneys are so damaged that they can't
nutritional health, acid-base abnormalities, and filter blood as well as they should. The kidneys' main
neurological complications. Early and error-free job is to get rid of waste and extra water from the
detection of CKD can help avert further deterioration blood.8 This is how urine is made. CKD means that
of a patient's health. These chronic diseases are waste has built up in the body. This condition is
prognosticated using various data mining chronic because the damage happens slowly over a
classification approaches and machine learning (ML) long period. It is a disease that affects people all over
algorithms. This Prediction is performed using the world.7 Because of CKD, you might experience
Logistic Regression, KNN, Random Forest, Decision various difficulties with your health. Diabetes, high
Tree, SVM, Gradient Boosting, Xgboost, Adaboost, blood pressure, and heart disease are only 3 of the
Naive Bayes, Bagged Decision Trees (Bagging) many conditions that can lead to CKD. In addition to
Classifier, and Voting Classifier. The data is these severe health problems, age and gender also
collected from the UCI Repository, which has 400 play a role in who gets CKD.26 If one or both of your
data sets with 21 attributes. This data has been fed kidneys aren't working right, you may have back
into Classification algorithms. The experimental pain, stomach pain, diarrhea, fever, nosebleeds, rash,
results show that DT, RF, and Gradient Boosting and vomiting. The two most common illnesses that
hands out an accuracy of 98.75%, 98.75% and might cause long-term damage to the kidneys are
97.50%, respectively. The Xgboost and Adaboost diabetes and high blood pressure.28 Therefore, the
classifier gives out a maximum accuracy of 100%. prevention of CKD can be thought of as the control
of these two diseases. Because chronic kidney disease
Keywords:- Chronic Kidney Disease (CKD), Early
(CKD) does not often present any symptoms until it
Diagnosis, Characterization, Treatment,
has progressed to a more advanced state, many
Hypertension, Anemia, Mineral Bone Disorder, Poor
people who have it do not realize they have it until it
Nutritional Health, Acid-Base Abnormalities,
is too late.
Neurological Complications, Data Mining,
Classification, Machine Learning, Logistic Chronic Kidney Disease (CKD) represents a
Regression, KNN, Random Forest, Decision Tree, significant global health challenge, with its
prevalence steadily increasing over the years. This indicating the broader applications of ML beyond
condition substantially burdens healthcare systems diagnosis alone. These studies underscore ML
worldwide due to its associated morbidity, mortality, techniques' potential to revolutionize CKD
and economic costs. Epidemiological studies have management by enabling personalized treatment
highlighted the rising prevalence of CKD across strategies and improving patient outcomes.
diverse populations, underscoring the need for
This review aims to provide a comprehensive
practical diagnostic and predictive approaches to
overview of recent CKD diagnosis and prognosis
mitigate its impact.
advancements, focusing on applying ML techniques.
Several studies have contributed to our understanding By synthesizing findings from critical studies in the
of CKD diagnosis and management. Zhang et al. [3] field, this review aims to elucidate the evolving
conducted a cross-sectional survey in China, landscape of CKD management and the pivotal role
revealing valuable insights into the prevalence of of ML in shaping its future trajectory.
CKD in this populous nation. Similarly, Singh et al.
LITERATURE SURVEY
[4] demonstrated the importance of incorporating
temporal electronic health record (EHR) data into Chronic Kidney Disease (CKD) is a significant
predictive models to stratify the risk of renal function global health issue, with its prevalence increasing
deterioration. These studies underscore the over the years. Researchers have explored various
multifaceted nature of CKD diagnosis, emphasizing methodologies to address this challenge, including
the necessity of leveraging advanced methodologies fuzzy classifiers, random forests, support vector
for accurate assessment and prognosis. machines (SVM), and machine learning (ML)
techniques, to improve CKD diagnosis, risk
Machine learning (ML) techniques have emerged as
stratification, and treatment response prediction.
powerful tools in healthcare, offering the potential to
enhance diagnostic accuracy and predictive Chen et al. [1] proposed using fuzzy classifiers for
capabilities. Researchers have explored various ML diagnosing CKD, showcasing the potential of fuzzy
algorithms to diagnose CKD and predict its logic in handling uncertainty in medical data. Subasi
progression. For instance, Subasi et al. [2] employed et al. [2] introduced random forest algorithms for
random forest algorithms for CKD diagnosis, CKD diagnosis, demonstrating its efficacy in
showcasing the utility of ML in this domain. handling large datasets and complex decision-making
Additionally, Polat et al. [6] demonstrated the processes.
efficacy of support vector machine algorithms
coupled with feature selection methods for CKD To understand the epidemiology of CKD, Zhang [3]

diagnosis, further highlighting the versatility of ML conducted a cross-sectional survey in China,

approaches. providing valuable insights into the prevalence and

distribution of CKD within the population. Singh et
Moreover, Barbieri et al. [7] introduced a novel ML al. [4] incorporated temporal Electronic Health
approach for predicting treatment response in end-
stage renal disease patients undergoing dialysis,
Record (EHR) data into predictive models, enhancing Additionally, ML has been utilized to handle clinical
risk stratification for renal function deterioration. text data. Du et al. [13] developed an ML-based
approach to identify protected health information in
Cueto-Manzano et al. [5] conducted a study on the
Chinese clinical text, addressing privacy concerns in
prevalence of CKD in an adult population,
healthcare data management. Abbas et al. [14]
contributing to understanding CKD burden in
employed ML techniques for classifying fetal distress
different demographics. Polat et al. [6] employed
and hypoxia, showcasing the applicability of ML in
SVM with feature selection methods for CKD
obstetrics.
diagnosis, highlighting the importance of selecting
relevant features to improve model performance. Moreover, Mahyoub et al. [15] compared ML
algorithms to rank Alzheimer's disease risk factors,
Barbieri et al. [7] proposed a novel ML approach for
highlighting the broader impact of ML in
predicting the response to anaemia treatment in end-
neurological research.
stage renal disease patients undergoing dialysis,
demonstrating the potential of ML in personalized In conclusion, the literature reflects various
medicine within nephrology. Papademetriou et al. [8] methodologies and approaches employed in CKD
investigated the relationship between CKD, basal diagnosis, epidemiology, risk stratification, and
insulin glargine, and health outcomes, emphasizing treatment response prediction. ML techniques, in
the importance of managing CKD in individuals with particular, have shown promise in improving CKD
dysglycemia. management and advancing personalized healthcare.
Further research in this domain promises to enhance
Hill [9] conducted a systematic review and meta-
our understanding and management of CKD, thereby
analysis to determine the global prevalence of CKD,
improving patient outcomes and reducing the burden
consolidating existing knowledge on CKD
of this chronic condition globally.
epidemiology. Hossain et al. [10] introduced
mechanical anisotropy assessment in the kidney II. METHODOLOGY
cortex, providing insights into tissue properties
Modules:
relevant to CKD progression.

 Data exploration: using this module we will

ML techniques have been applied to various domains
load data into system
outside the realm of nephrology. Alloghani et al. [11]
explored ML applications in software engineering
 Processing: Using the module we will read data
learning, showcasing the versatility of ML
for processing
methodologies. Gupta et al. [12] proposed a method
to predict diagnostic codes for chronic diseases,  Splitting data into train & test: using this
demonstrating the potential of ML in healthcare module data will be divided into train & test
management beyond CKD.
 Model generation: Model building - Logistic
Regression, KNN, Random Forest, Decision
 Tree, SVM, Gradient Boosting, Xgboost, Xgboost and Adaboost classifier gives out a
Adaboost, Naive Bayes, Bagged Decision Trees maximum accuracy of 100%.
(Bagging) Classifier, and Voting Classifier B) Dataset Collection
The dataset used in this study comprises 400
 User signup & login: Using this module will get
instances gathered from the UCI Repository,
registration and login
containing 21 attributes related to chronic kidney
disease (CKD). These attributes likely encompass
 User input: Using this module will give input
crucial indicators such as patient demographics,
for prediction
medical history, laboratory test results, and
 Prediction: final predicted displayed symptomatology. Each instance represents a unique
patient case with various features pertinent to CKD
A) System Architecture diagnosis and prognosis. Attributes may include
demographic information (e.g., age, gender), clinical
measurements (e.g., blood pressure, serum creatinine
levels), laboratory test results (e.g., urinalysis
findings), and symptoms associated with CKD (e.g.,
fatigue, oedema). The dataset aims to capture a
comprehensive overview of patient profiles afflicted
with CKD, enabling the application of data mining
classification approaches and machine learning
algorithms for predictive analysis. The dataset's
Fig 1: System Architecture richness allows for robust model training and
evaluation, facilitating the identification of patterns
Proposed work
and relationships essential for accurate CKD
Early and error-free detection of CKD can help avert
prediction. This diverse dataset empowers researchers
further deterioration of a patient's health. These
and practitioners in the healthcare domain to develop
chronic diseases are prognosticated using various
practical diagnostic and prognostic tools for early
data mining classification approaches and machine
detection and management of CKD, thereby
learning (ML) algorithms. This Prediction uses
mitigating potential health complications and
Logistic Regression, KNN, Random Forest, Decision
improving patient outcomes.
Tree, SVM, Gradient Boosting, Xgboost, Adaboost,
and Ensemble. The data is collected from the UCI
C) Pre-processing
Repository, which has 400 data sets with 21
attributes. This data has been fed into Classification To prepare the dataset for classification algorithms
algorithms. The experimental results show that DT, such as Logistic Regression, KNN, Random Forest,
RF, and Gradient Boosting hands out an accuracy of Decision Tree, SVM, Gradient Boosting, XGBoost,
98.75%, 98.75% and 97.50%, respectively. The Adaboost, Naive Bayes, Bagged Decision Trees
(Bagging) Classifier, and Voting Classifier, the D) Training & Testing
following preprocessing steps can be applied:
In this study, the training and testing process
Handling Missing Values: Check for missing data in involved utilizing various machine learning
the dataset and employ strategies like imputation classifiers to predict chronic kidney disease (CKD)
(replacing missing values with the mean, median, or based on a dataset obtained from the UCI Repository.
mode of the respective attribute) or deletion of Initially, the dataset consisting of 400 instances with
instances/attributes with missing values. 21 attributes was divided into two subsets: a training
set and a testing set. The training set, which
Normalization/Standardization: Scale the numeric
comprised most of the data, was used to train the
features to a uniform range to prevent any particular
classifiers, while the testing set was kept separate for
feature from dominating the others. This ensures that
evaluating the performance of the trained models.
the algorithms treat all features equally. Methods like
Min-Max scaling or Z-score normalization can be For training, the classifiers, including Logistic
applied. Regression, KNN, Random Forest, Decision Tree,
SVM, Gradient Boosting, Xgboost, Adaboost, Naive
Feature Selection/Dimensionality Reduction:
Bayes, Bagged Decision Trees (Bagging) Classifier,
Analyze the relevance of each attribute and remove
and Voting Classifier, were employed. Each classifier
irrelevant or redundant features to improve model
underwent a training phase where it learned patterns
performance and reduce computation time.
and relationships within the training data to make
Techniques like Principal Component Analysis
predictions about CKD.
(PCA) or feature importance from tree-based models
can be employed. Following the training phase, the trained models were
evaluated using the testing set to assess their
Handling Categorical Variables: Encode categorical
predictive performance. The accuracy of each
variables into numerical format using techniques like
classifier was calculated based on its ability to
one-hot encoding or label encoding, depending on the
classify instances of CKD within the testing set
nature of the data and the algorithm requirements.
correctly. Experimental results revealed that Decision

Balancing Data (if needed): If there's a class Tree, Random Forest, and Gradient Boosting

imbalance issue, techniques like oversampling classifiers achieved accuracies of 98.75%. In

(replicating instances of the minority class) or comparison, Xgboost and Adaboost classifiers

undersampling (removing instances of the majority achieved a maximum accuracy of 100%, indicating

class) can be applied to balance the dataset. their effectiveness in predicting CKD.

By performing these preprocessing steps, the dataset Algorithms.

will be well-prepared for training classification

Logistic Regression:
algorithms, ensuring optimal performance and
accurate predictions for early diagnosis and Logistic Regression is a supervised machine learning
characterization of chronic kidney disease. algorithm used for binary classification tasks. Using a
logistic function, it models the probability that a individual trees' mean Prediction (Regression). It
given input belongs to a particular category. Despite reduces over fitting by averaging multiple trees.
its name, it's used for classification, not Regression. It
works well when the relationship between the
dependent and independent variables is linear or can
be transformed into a linear form.

Decision Tree:

A Decision Tree is a tree-like structure where internal

nodes represent feature tests, branches represent
KNN (K-Nearest Neighbors):
decisions, and leaf nodes represent outcomes. It's a
KNN is a simple yet effective supervised learning popular algorithm for classification and regression
algorithm for classification and regression tasks. It tasks due to its simplicity and interpretability.
classifies new data points based on the majority class
of their k-nearest neighbors in the feature space.
KNN's performance heavily relies on the choice of
distance metric and the value of k.

SVM (Support Vector Machine):

SVM is a robust supervised learning algorithm used

for classification and regression tasks. It finds the

Random Forest: hyper plane that best separates classes by maximizing

the margin between them.
Random Forest is an ensemble learning method that
constructs many decision trees during training and
outputs the mode of the classes (classification) or the
 Adaboost:

Adaboost (Adaptive Boosting) is an ensemble

learning method that combines multiple weak
classifiers to build a strong classifier. It adjusts the
weights of misclassified instances to focus more on
difficult cases in subsequent iterations.

Gradient Boosting:

Gradient Boosting is an ensemble learning technique

that builds models sequentially, where each model
corrects the errors of its predecessor. It minimizes a
loss function by adding weak learners (usually Naive Bayes:
decision trees) in a stage-wise manner.
Naive Bayes is a probabilistic classifier based on
Bayes' theorem with the "naive" assumption of
independence between features. Despite its
simplicity, it often performs well in practice,
especially for text classification tasks.

XGBoost:

XGBoost (Extreme Gradient Boosting) is an

optimized implementation of gradient boosting
designed for speed and performance. It employs
regularization techniques to prevent overfitting and
Bagged Decision Trees (Bagging) Classifier:
can handle missing values efficiently.

Bagging is an ensemble meta-algorithm that

improves the accuracy and stability of machine
learning algorithms by training multiple models on
different subsets of the training data and combining
their predictions through averaging or voting.
 To quantify a test's exactness, we should register the
negligible part of genuine positive and genuine
adverse outcomes in completely examined cases.
This might be communicated numerically as:

Accuracy = TP + TN TP + TN + FP + FN.

Voting Classifier:
Precision: Precision measures the proportion of
A Voting Classifier combines the predictions of properly categorized occurrences or samples among
multiple individual models (classifiers or regressors) the positives. As a result, the accuracy may be
and predicts the class with the highest majority vote calculated using the following formula:
(for classification) or averages the predictions (for
regression). It can be hard or soft voting, depending Precision = True positives/ (True positives + False

on how the individual models' outputs are combined. positives) = TP/(TP + FP)

It often yields better performance than individual

models.

Recall: Recall is a machine learning metric that

surveys a model's capacity to recognize all pertinent
examples of a particular class. It is the proportion of
appropriately anticipated positive perceptions to add
up to real up-sides, which gives data about a model's
capacity to catch instances of a specific class.

III. EXPERIMENTAL RESULTS

A) Comparison Graphs → Accuracy, Precision,

Recall, f1 score F1-Score: The F1 score is a machine learning
evaluation measurement that evaluates the precision
Accuracy: A test's accuracy is defined as its ability of a model. It consolidates a model's precision and
to recognize debilitated and solid examples precisely. review scores. The precision measurement computes
how often a model anticipated accurately over the full
dataset.

Fig 4: Home page

B) Performance Evaluation Graph

Fig 5: User Signup page

Fig 6: User Sign in Page

Fig 2: Performance Evaluation Graph

C) Frontend

Fig 7: Sample data for testing

Fig 3: Url Link to Web Page

 models and exploring explainable AI methods can
enhance trust and adoption in clinical settings,
ultimately improving patient care.

V. FUTURE SCOPE

Future advancements in CKD prediction models

Fig 8: Result: you are safe! No Disease could involve integrating advanced deep learning
techniques like neural networks to capture intricate
IV. CONCLUSION
data patterns missed by traditional algorithms. Real-

Early and error-free detection of CKD can help avert time patient data integration and continuous

further deterioration of a patient's health. These monitoring technologies could enhance model

chronic diseases are prognosticated using various responsiveness to dynamic health changes.

data mining classification approaches and machine Collaboration with healthcare professionals enables

learning (ML) algorithms. This Prediction uses the inclusion of domain-specific features, ensuring

Logistic Regression, KNN, Random Forest, Decision practicality and clinical relevance. Addressing model

Tree, SVM, Gradient Boosting, Xgboost, Adaboost, interpretability through explainable AI methods will

and Ensemble. The data is collected from the UCI enhance trust and adoption in clinical settings,

Repository, which has 400 data sets with 21 ultimately improving patient care. Further research

attributes. This data has been fed into Classification could explore personalized medicine approaches and

algorithms. The experimental results show that DT, predictive analytics for early intervention, paving the

RF, and Gradient Boosting hands out an accuracy of way for more effective CKD management and patient

98.75%, 98.75% and 97.50%, respectively. The outcomes.

Xgboost and Adaboost classifier gives out a

REFERENCES
maximum accuracy of 100%.
[1] Z. Chen, Z. Zhang, R. Zhu, Y. Xiang, and P. B.
Future research could explore integrating advanced
Harrington, ``Diagnosis of patients with chronic
deep learning techniques, such as neural networks, to
kidney disease by using two fuzzy classifiers,''
advance CKD prediction models further and capture
Chemometrics Intell. Lab. Syst., vol. 153, pp.
intricate patterns within the data that traditional
140145, Apr. 2016.
algorithms might miss. Additionally, incorporating
real-time patient data and leveraging continuous [2] A. Subasi, E. Alickovic, and J. Kevric,
monitoring technologies could enhance the model's ``Diagnosis of chronic kidney disease by using
responsiveness to dynamic health changes. random forest,'' in Proc. Int. Conf. Med. Biol.
Collaborations with healthcare professionals may Eng.,Mar. 2017, pp. 589594.
enable the inclusion of domain-specific features and
[3] L. Zhang, ``Prevalence of chronic kidney disease
insights, ensuring the model's practicality and clinical
in China: A crosssectionalsurvey,'' Lancet, vol. 379,
relevance. Addressing the interpretability of the
pp. 815822, Mar. 2012.
[4] A. Singh, G. Nadkarni, O. Gottesman, S. B. Ellis, Bellinger,L. E. Wimsey, and C. M. Gallippi,
E. P. Bottinger, andJ. V. Guttag, ``Incorporating ``Mechanical anisotropy assessment in kidney cortex
temporal EHR data in predictive models for risk using ARFI peak displacement: Preclinical validation
stratication of renal function deterioration,'' J. and pilot in vivo clinical results in kidney allografts,''
Biomed. Informat.,vol. 53, pp. 220228, Feb. 2015. IEEE Trans. Ultrason.,Ferroelectr., Freq. Control,
vol. 66, no. 3, pp. 551562, Mar. 2019.
[5] A. M. Cueto-Manzano, L. Cortés-Sanabria, H. R.
Martínez-Ramírez,E. Rojas-Campos, B. Gómez- [11] M. Alloghani, D. Al-Jumeily, T. Baker, A.
Navarro, and M. Castillero-Manzano,``Prevalence of Hussain, J. Mustana, andA. J. Aljaaf, ``Applications
chronic kidney disease in an adult population,'' Arch. of machine learning techniques for software
Med.Res., vol. 45, no. 6, pp. 507513, Aug. 2014. engineering learning and early prediction of students'
performance,'' in Proc. Int. Conf. Soft Comput. Data
[6] H. Polat, H. D. Mehr, and A. Cetin, ``Diagnosis
Sci., Dec. 2018, pp. 246258.
of chronic kidney disease based on support vector
machine by feature selection methods,'' J. Med.Syst., [12] D. Gupta, S. Khare, and A. Aggarwal, ``A
vol. 41, no. 4, p. 55, Apr. 2017. method to predict diagnostic codes for chronic
diseases using machine learning techniques,'' in
[7] C. Barbieri, F. Mari, A. Stopper, E. Gatti, P.
Proc.Int. Conf. Comput., Commun. Autom.(ICCCA),
Escandell-Montero,J. M. Martínez-Martínez, and J.
Apr. 2016, pp. 281287.
D. Martín-Guerrero, ``A new machine learning
approach for predicting the response to anemia [13] L. Du, C. Xia, Z. Deng, G. Lu, S. Xia, and J.
treatment in a large cohort of end stage renal disease Ma, ``A machine learning based approach to identify
patients undergoing dialysis,'' Comput.Biol. Med., protected health information in Chinese clinical text,''
vol. 61, pp. 5661, Jun. 2015. Int. J. Med. Informat., vol. 116, pp. 2432, Aug. 2018.

[8] V. Papademetriou, E. S. Nylen, M. Doumas, J. [14] R. Abbas, A. J. Hussain, D. Al-Jumeily, T.

Probsteld, J. F. Mann,R. E. Gilbert, and H. C. Baker, and A. Khattak, ``Classification of foetal
Gerstein, ``Chronic kidney disease, basal insulin distress and hypoxia using machine learning
glargine, and health outcomes in people with approaches,''in Proc. Int. Conf. Intell.Comput., Jul.
dysglycemia: The ORIGIN Study,'' Amer. J. Med., 2018, pp. 767776.
vol. 130, no. 12, pp. 1465.e271465.e39, Dec. 2017.
[15] M. Mahyoub, M. Randles, T. Baker, and P.
[9] N. R. Hill, ``Global prevalence of chronic kidney Yang, ``Comparison analysis of machine learning
disease A systematic review and meta-analysis,'' algorithms to rank alzheimer's disease risk factors by
PLoS ONE, vol. 11, no. 7, Jul. 2016,Art. no. importance,'' in Proc. 11th Int. Conf. Develop. eSyst.
e0158765. Eng. (DeSE),Sep. 2018, pp. 111.

[10] M. M. Hossain, R. K. Detwiler, E. H. Chang, M.

C. Caughey, M.W. Fisher,T. C. Nichols, E. P.
Merricks, R. A. Raymer, M. Whitford, D. A.