# Detecting Chronic Kidney Disease(CKD) at the Initial Stage: A Novel Hybrid Feature-selection Method and Robust Data Preparation Pipeline for Different ML Techniques

**Authors:** Md. Taufiqul Haque Khan Tusar, Md. Touhidul Islam, Foyjul Islam Raju  
**arXiv:** [2203.01394](https://arxiv.org/abs/2203.01394)  
**Date:** 2022-03-02  
**Summary page:** https://www.summarizepaper.com/en/paper/2203.01394v1/

## Key points

- Chronic Kidney Disease (CKD) affects almost 800 million people globally, with 1.7 million deaths annually.
- Early detection of CKD is crucial for saving lives.
- Researchers have applied various Machine Learning (ML) methods to detect CKD at an early stage, but detailed studies are still lacking in this area.
- The authors present a structured and comprehensive method for dealing with medical data complexities with optimal performance.
- The proposed method includes KNN Imputation, Local Outlier Factor, SMOTE, K-stratified K-fold Cross-validation, and a novel hybrid feature selection method to remove redundant features.
- The applied algorithms include Support Vector Machine, Gaussian Naive Bayes, Decision Tree, Random Forest, Logistic Regression, K-Nearest Neighbor, Gradient Boosting, Adaptive Boosting and Extreme Gradient Boosting.
- The authors' approach achieved excellent results as the Random Forest algorithm detected CKD with 100% accuracy without any data leakage.
- This study's findings provide valuable insights into developing effective ML models for detecting CKD at an early stage and can contribute significantly to improving patient outcomes globally.

## Summary

Chronic Kidney Disease (CKD) is a global health issue that affects almost 800 million people, with 1.7 million deaths annually. Early detection of CKD is crucial for saving lives, and many researchers have applied various Machine Learning (ML) methods to detect it at an early stage. However, detailed studies are still lacking in this area. In this paper, the authors present a structured and comprehensive method for dealing with medical data complexities with optimal performance. The study aims to assist researchers in producing clear ideas on the medical data preparation pipeline. The proposed method includes KNN Imputation for imputing missing values, Local Outlier Factor for removing outliers, SMOTE for handling data imbalance, K-stratified K-fold Cross-validation for validating ML models, and a novel hybrid feature selection method to remove redundant features. The applied algorithms include Support Vector Machine, Gaussian Naive Bayes, Decision Tree, Random Forest, Logistic Regression, K-Nearest Neighbor, Gradient Boosting, Adaptive Boosting and Extreme Gradient Boosting. The authors' approach achieved excellent results as the Random Forest algorithm detected CKD with 100% accuracy without any data leakage. This study's findings provide valuable insights into developing effective ML models for detecting CKD at an early stage and can contribute significantly to improving patient outcomes globally.