A Comparative Analysis of Machine Learning Approaches for Chronic Kidney Disease Detection

Abstract

In performing tests and diagnosing illness in healthcare, symptom experiment processes are used to detect the fundamental cause of syndromes. In this research, we employed a crossbred technique to develop our mentioned ideal model, which improvd Pearson correlation for the purpose of feature selection. The initial phase involved choosing the ideal models through critical analysis of the literature. Subsequently, our proposed hybrid model incorporated a combination of these models. The base classifiers used in XGBoost, Random Forest, Logistic Regression, AdaBoost and Hybrid model classifiers, while the Meta classifier was the random forest classifier. Mainly the goal of this experiment is to assess the most useful machine learning classification methods and predict the optimal classifier in terms of precision. This makes amends for the problem of overfitting and attains the best level of accuracy. The focus of the evaluation is on accuracy, and a detailed analysis of the relevant literature is presented in a tabular format. To implementation, we used four best-performing ML models and developed a new model named hybrid utilizing the UCI Chronic Kidney Failure dataset for predictive purposes. An XGBoost classifier achieves about 98% accuracy, a random forest achieves 97% accuracy, Logistic Regression about 94% accuracy, AdaBoost achieves 95% accuracy, and our proposed new model named hybrid achieves the highest 99% accuracy, and it performs best on the same dataset. Several prominent machine learning models utilized to predict the onset of CKF (chronic kidney failure) include Naïve Bayes, Random Forest, Decision Tree, Support Vector Machine, K-nearest neighbor, LDA, GB, and neural network. Within our research, XGBoost, AdaBoost, Logistic Regression, random forest, and hybrid models are used for the same dataset of features for comparison of accuracy scores.