Comparative analysis of weka-based classification algorithms on medical diagnosis datasets

Area of Science:

• Medical Informatics
• Machine Learning
• Data Science

Background:

• The proliferation of Big Data and 5G technology has led to a surge in medical data from electronic health records and digitized equipment.
• Hospitals now possess vast databases containing clinical diagnosis and hospital management information.
• Advancements in medical information technology necessitate efficient data analysis methods.

Purpose of the Study:

• To evaluate the classification performance of diverse machine learning algorithms on medical datasets.
• To explore the potential of machine learning in enhancing medical diagnosis accuracy.
• To identify optimal algorithms for specific medical data classification tasks.

Main Methods:

• Utilized four distinct medical classification datasets from the University of California Irvine machine learning repository.
• Developed six categories of classification models using the Weka platform, based on Bayesian, ensemble, rule-based, and tree-based approaches.
• Conducted between-group and within-group experiments to compare algorithm performance.

Main Results:

• Random Forest algorithm exhibited the highest accuracy on the Indian Liver Patient Dataset (ILPD), Cardiotocography (CADG), and Lymphatic Tractography (LYMP) datasets.
• Bagging algorithm outperformed other ensemble methods across multiple metrics, particularly on binary datasets.
• Logistic Model Tree achieved optimal results on the Mammographic dataset (MAGR), while Random Forest excelled on ILPD, CADG, and LYMP.

Conclusions:

• Machine learning algorithms possess significant potential for disease prediction and diagnostic support.
• The findings provide a valuable reference for applying machine learning in clinical decision-making.
• Specific algorithms like Random Forest and Bagging show strong applicability in medical data classification.