“Detecting Kidney Disease Before It’s Too Late”

“Detecting Kidney Disease Before It’s Too Late”

Machine Learning for Early Detection of Chronic Kidney Disease

Under Guidance of : Prof. S. D. Khatavkar

Academic Year 2025–2026

Detecting Kidney Disease Before It’s Too Late

Medium Link: Detecting Kidney Disease

Executive Summary

This project, “Detecting Kidney Disease Before It’s Too Late,” focuses on using machine learning techniques to enable early detection of chronic kidney disease (CKD), a condition that often progresses silently and becomes severe if not diagnosed in time. By analyzing patient data from sources like the UCI CKD dataset [6], machine learning models such as Decision Trees and Random Forest can identify patterns and predict the likelihood of CKD with high accuracy, supporting faster and more reliable diagnosis [1], [2]. These approaches have shown strong performance in medical studies, helping clinicians make better decisions based on data-driven insights. Additionally, techniques like feature analysis and explainable AI improve understanding of key health indicators influencing the disease [5]. Overall, this project aims to assist healthcare professionals in early intervention, reduce disease progression, and improve patient outcomes.

Introduction

Chronic Kidney Disease (CKD) is a progressive condition in which the kidneys gradually lose their ability to function properly. Because its early stages often show little to no symptoms, many patients are diagnosed only after significant damage has occurred. This makes early detection essential to prevent serious complications and improve overall patient health outcomes.

With the advancement of Machine Learning (ML), it has become possible to analyze medical data and predict diseases like CKD at an early stage. Various studies have shown that ML algorithms such as Decision Trees, Random Forest, and Support Vector Machines can effectively predict CKD using datasets like the UCI CKD dataset [1], [2], [6]. Additionally, modern techniques like SHAP improve model interpretability by identifying key factors influencing the disease [5]. This project, “Detecting Kidney Disease Before It’s Too Late,” focuses on using these techniques to support early diagnosis and better healthcare decisions.

A Brief History of the CKD

Chronic Kidney Disease (CKD) has long been recognized as a major global health issue, traditionally diagnosed through laboratory tests and clinical observation. In earlier years, detection relied heavily on symptoms and basic medical reports, which often appeared only in the later stages of the disease. This delay in diagnosis led to increased complications, higher treatment costs, and reduced chances of recovery, highlighting the need for earlier and more accurate detection methods.

With the advancement of digital healthcare and data analysis, researchers began exploring the use of Machine Learning (ML) to improve early diagnosis of CKD. Over time, studies demonstrated that ML algorithms could analyze patient data more efficiently and predict the disease at an earlier stage using datasets like the UCI CKD dataset [6]. Recent research has further enhanced these models by incorporating advanced techniques and explainable AI methods, making predictions more accurate and reliable [1], [2], [5]. This evolution reflects a shift from traditional diagnosis to data-driven approaches aimed at detecting CKD before it becomes severe.

Figure 1: Structure of Human Kidney

Existing Solutions and Their Limitations

Before the use of machine learning, chronic kidney disease (CKD) was mainly diagnosed using traditional medical methods such as blood tests, urine analysis, and clinical observation. Doctors relied on parameters like creatinine levels, blood pressure, and patient history to identify the disease. While these methods are reliable, they often detect CKD only in later

stages, as early symptoms are minimal or no noticeable. This delay reduces the chances of effective treatment and increases the risk of complications.

With the introduction of early computational approaches, basic statistical models and simple machine learning algorithms were used to predict CKD using datasets like the UCI CKD dataset [6]. Studies showed that models such as Decision Trees and Support Vector Machines could improve prediction accuracy [1], [2]. However, these solutions still had limitations such as handling missing or incomplete data, lower accuracy in some cases, lack of interpretability, and dependency on limited datasets. Additionally, many models were not user-friendly or easily applicable in real clinical settings. These limitations highlight the need for more advanced, accurate, and explainable solutions, which this project aims to address [5].

Algorithm Comparison and Justification

Feature

Decision Tree

Random Forest

Support Vector Machine (SVM)

Working Principle

Uses a single tree with decision rules

Uses multiple trees

Uses hyperplane to separate data classes

Accuracy

Moderate

High

High

Overfitting

High

Low

Moderate

Interpretability

Easy to understand and visualize

Moderate (harder than a single tree)

Difficult

Training Speed

Fast

Slower than Decision Tree

Slow for large datasets

Performance

Less stable (high variance)

More stable and reliable

Good but sensitive to parameters

Why Random Forest is Preferred?

Among the compared algorithms, Random Forest provides better performance for CKD prediction due to its ensemble nature. It combines multiple decision trees and reduces overfitting by averaging their predictions.

Unlike Decision Trees, which can become overly complex and memorize training data, Random Forest generalizes better to unseen data. Compared to SVM, Random Forest is easier to implement, requires less parameter tuning, and performs well even with missing or noisy data.

Therefore, Random Forest is selected as the primary model in this project due to its higher accuracy, robustness, and reliability in medical prediction tasks.

Proposed Solution

To overcome the limitations of existing methods, this project proposes a machine learning-based system for early detection of Chronic Kidney Disease (CKD). The solution uses patient health data from the UCI CKD dataset [6] and applies advanced algorithms such as Random Forest and Decision Tree to accurately predict whether a person is likely to have CKD. These models are trained to identify hidden patterns in the data, enabling early diagnosis even before severe symptoms appear.

Data Preprocessing and Model Pipeline

a. Data Preprocessing

Before training the machine learning models, the dataset undergoes several preprocessing steps to ensure data quality and improve model performance.

b. Handling Missing Values:

The UCI CKD dataset contains missing values in several attributes. These are handled using appropriate techniques such as mean/median imputation for numerical features

and mode imputation for categorical features.

c. Encoding Categorical Data:

Some features such as red blood cells, hypertension, and diabetes are categorical. These are converted into numerical form using label encoding or one-hot encoding so that machine learning algorithms can process them.

d. Feature Scaling (Normalization):

To ensure that all features contribute equally to the model, numerical values are normalized or standardized using techniques like Min-Max scaling or StandardScaler. These preprocessing steps help improve the accuracy, stability, and efficiency of the model.

Figure 2: Proposed Machine Learning Framework for CKD Prediction

In addition, the system incorporates feature analysis techniques like SHAP to improve transparency and help understand which health factors (such as creatinine, blood pressure, etc.) play a major role in prediction [5]. The proposed solution is designed to be more accurate, reliable, and user-friendly compared to earlier approaches [1], [2]. It can assist doctors in making faster decisions and support preventive healthcare by detecting the disease at an early stage.

Figure 2: Confusion Matrix for CKD Prediction Model

Mathematical Foundations of the Model

1. Accuracy Formula

𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦=𝑇𝑃+𝑇𝑁𝐹𝑃+𝐹𝑁𝑇𝑃+𝑇𝑁

Accuracy measures the overall correctness of the model by calculating the ratio of correctly predicted instances (both positive and negative) to the total number of predictions.

2. Precision Formula

Precision = 𝑇𝑃𝑇𝑃+𝐹𝑃

Precision indicates how many of the predicted positive cases (CKD detected) are actually correct. It is important when false positives need to be minimized.

3. Gini Index (Decision Tree)

Gini=1−Σ𝑝𝑖2𝑛𝑖=1

The Decision Tree algorithm uses the Gini Index to decide the best feature for splitting the data. A lower Gini value indicates a purer node, meaning better separation of classes.

Figure 3: ROC Curve Showing Model Performance

REFERENCES:

[1] G. Nandhini and J. Aravinth, Chronic Kidney Disease Prediction Using Machine Learning Technique, IEEE Conference, 2021. https://doi.org/10.1109/RTEICT52294.2021.9573971

[2] D. A. Debal and T. M. Sitote, Chronic Kidney Disease Prediction Using Machine Learning Techniques, Journal of Big Data, 2022. https://journalofbigdata.springeropen.com/articles/10.1186/s40537-022-00657-5

[3] Supriya G. et al., Prediction of Chronic Kidney Disease Using Machine Learning, IJARCCE, 2024. https://ijarcce.com/papers/prediction-of-chronic-kidney-disease-using-machine-learning/

[4] V. Srilakshmi et al., Chronic Kidney Disease Prediction Based on Machine Learning Algorithms, IJITCE, 2025. https://ijitce.org/index.php/ijitce/article/view/1189

[5] M. Islam and R. A. Ripon, Prediction of Chronic Kidney Disease with Machine Learning Models and Feature Analysis Using SHAP, RIA Journal, 2023. https://www.iieta.org/journals/ria/paper/10.18280/ria.370226

[6] UCI Machine Learning Repository, Chronic Kidney Disease Dataset. https://archive.ics.uci.edu/ml/datasets/chronic_kidney_disease

[7] Rahman et al., Machine learning models for chronic kidney disease diagnosis, 2024.

[8] Ghosh et al., Explainable ML for CKD prediction, 2024.