Lisa Ariyanti, Alamsyah Alamsyah

C4.5 Algorithm Optimization and Support Vector Machine by Applying Particle Swarm Optimization for Chronic Kidney Disease Diagnosis

Introduction

C4.5 algorithm optimization and support vector machine by applying particle swarm optimization for chronic kidney disease diagnosis. Optimize Chronic Kidney Disease (CKD) diagnosis using C4.5 & Support Vector Machine (SVM) algorithms with Particle Swarm Optimization (PSO). Achieves 100% accuracy for C4.5 and 98.75% for SVM.

Abstract

Abstract. Kidneys are one of the organs of the body that have a very important function in life. The main function of the kidneys is to excrete metabolic waste products. Chronic kidney disease is a result of the gradual loss of kidney function. Chronic kidney disease occurs when the kidneys are unable to maintain an internal environment consistent with life and the restoration of useless functions. Data mining is one of the fastest growing technologies in biomedical science and research. Purpose: In the field of medicine, data mining can improve hospital information management and telemedicine development. In the first stage of data mining process, data processing is done with pre-processing by handling missing values ​​and data transformation. Then, the feature selection stage is carried out using the Particle Swarm Optimization algorithm to find the best attributes. Next, it is done by classifying the dataset. Methods/Study design/approach: The algorithm used for classification is the C4.5 Algorithm and the Support Vector Machine. Both classifications are known as algorithms that have a fairly good level of accuracy. This study uses the chronic kidney disease dataset from the UCI Machine Learning Repository. Result/Findings: This research increases the accuracy by 100% for the C4.5 Algorithm and 98.75% for the Support Vector Machine by using 24 attributes and 1 class attribute. Novelty/Originality/Value: The purpose of this study was to determine the level of accuracy of the comparison between the C4.5 Algorithm and the Support Vector Machine after applying the Particle Swarm Optimization algorithm. 

Review

This paper addresses the significant clinical challenge of chronic kidney disease (CKD) diagnosis using data mining techniques, a highly relevant and impactful area within biomedical research. The authors propose a methodology that integrates Particle Swarm Optimization (PSO) for feature selection, subsequently applying C4.5 and Support Vector Machine (SVM) algorithms for classification. The stated objective is to compare the accuracy of these classifiers after PSO-based feature selection. The most striking claim from the abstract is the achievement of 100% accuracy for the C4.5 algorithm and 98.75% for SVM, utilizing a subset of 24 attributes from the UCI CKD dataset. While the topic is pertinent and the general approach aligns with current trends in medical AI, these exceptionally high accuracy figures warrant a thorough and critical examination. The study's strengths lie in its clear objective of improving CKD diagnosis and its use of well-established machine learning algorithms. The methodology outlines a standard data mining pipeline, including pre-processing for missing values and data transformation, followed by feature selection using PSO to identify optimal attributes. This sequential approach of enhancing the input space before classification is a sound strategy to potentially improve model performance. The choice of C4.5 and SVM, both recognized for their classification capabilities, provides a robust baseline for comparison. Furthermore, the use of the publicly available UCI Machine Learning Repository's chronic kidney disease dataset allows for potential reproducibility and direct comparison with other research efforts in the domain. Despite the promising reported results, several critical concerns and areas for clarification need to be addressed. Primarily, the claim of 100% accuracy for C4.5 is highly unusual for a real-world medical dataset and strongly suggests potential issues such as overfitting, data leakage, or an inadequate validation strategy (e.g., lack of proper cross-validation or a sufficiently independent test set). The abstract does not detail the validation methodology, which is crucial for substantiating such claims. Additionally, the term "C4.5 Algorithm Optimization" in the title is somewhat misleading, as PSO is applied for *feature selection* to benefit both C4.5 and SVM, rather than directly optimizing C4.5's internal parameters or structure. The abstract lacks specifics regarding the PSO implementation (e.g., swarm size, iterations, fitness function) and how the "best attributes" were truly identified. A more comprehensive comparative analysis, including performance metrics beyond just accuracy (e.g., sensitivity, specificity, F1-score) and comparison against other feature selection methods or models without PSO, would significantly strengthen the paper's contribution and provide a more nuanced understanding of the proposed approach's actual value.

Full Text

Comments

You need to be logged in to post a comment.