A Comprehensive Study for Predicting Chronic Kidney Disease, Diabetes, Hypertension, and Anemia by Machine Learning and Feature Engineering Techniques

Chronic Kidney Disease, Diabetes, Hypertension, and Anemia are affecting more people these days and causing serious deterioration in health conditions, which can cause death if left undiagnosed and untreated. Machine learning models can play an indispensable role in precisely predicting diseases at an early stage which can help doctors start the disease-management plan early and reduce the suffering of patients and the death rates. In this study, we propose machine learning based Chronic Kidney Disease, Diabetes, Hypertension, and Anemia Prediction. We analyzed Chronic_Kidney_Disease Data Set from the UCI repository. After data-prepossessing, we created four new datasets from the initial dataset for predicting the four diseases. We applied Feature Engineering on every dataset to identify the best features. We developed five machine learning based models and compared the models’ performance before and after Feature Engineering for every dataset. The Random Forest model performs best for chronic kidney disease prediction with an accuracy of 99.5%, validation score of 99.0%, and ROC-AUC score of 1.0. The Logistic Regression model gives the highest accuracy of 88.8%, validation score of 82.0%, and ROC-AUC score of 0.94 for predicting diabetes. For hypertension prediction, XGBoost outperforms other models with an accuracy of 88.8%, validation score of 83.2%, and ROCAUC score of 0.95. XGboost model best-predicted anemia with an accuracy of 88.8%, validation score of 91%, and ROC-AUC score of 0.91. Since the developed models can accurately perform these diseases’ predictions, we believe this study will be beneficial for the diagnosis and management of these diseases.