Accuracy and performance analysis for classification algorithms based on biomedical datasets

• Main Authors: Al-Hameli, Bassam Abdo, Alsewari, Abdulrahman A., Khubrani, Mousa, Fakhreldin, Mohammoud
• Format: Conference or Workshop Item
• Language: English
• Published: IEEE 2021
• Subjects: QA Mathematics
• Online Access: http://umpir.ump.edu.my/id/eprint/32649/1/Accuracy_and_performance_analysis_for_classification_algorithms_based_on_biomedical_datasets.pdf; http://umpir.ump.edu.my/id/eprint/32649/; https://doi.org/10.1109/ICSECS52883.2021.00119

Diseases chronic, including heart disease, cancer, diabetes, and obesity, are the main causes of mortality in the United States and accounting for and consuming the majority of the country’s healthcare expenditure. As indicated by recent researches. The main reason for the emergence of these diseases prominently is their relationship to each other, where diabetes is one of the causes of cancer and heart disease, hepatitis also is associated with diabetes, and heart disease. This paper focuses on data mining and machine learning techniques in healthcare classification and prediction of diseases and rebuild disease detection systems (DDS). The study suggests finding a classifier among the most common kinds of classification algorithms within a combined approach represent in Bayesian, Trees, Rules, Function, and lazy algorithms to automate a better performance of early detection of diseases from the medical datasets. This paper presents and analyzes five different machine learning (ML) algorithms: Function-based Neural Network (MLP) algorithm. Trees based Decision Tree (ID3) algorithm, Bayesian Theorem based Hidden Naïve Bayes (HNB) algorithm. Lazy based k-nearest neighbors (IBK) algorithm, and Rules-based OneR algorithm. The analysis is based on four benchmark datasets in the healthcare sector, including the Pima Indian Diabetes PID, the Breast Cancer, Heart Cleveland, and Hepatitis Datasets, which were obtained from several ML repositories. The results show that the HNB predicts the best result with a relatively higher Precision, AUROC Statistic, highest accuracy, and performance when compared to MLP, IBK, OneR, ID3 algorithms.