Related Experiment Videos
Inductive learning of thyroid functional states using the ID3 algorithm. The effect of poor examples on the learning
1Department of Clinical Chemistry, University of Turku, Finland.
International Journal of Bio-Medical Computing
|January 1, 1992
Summary
The ID3 algorithm for inductive learning showed improved accuracy with more thyroid patient data. However, it struggled with incomplete datasets and could overlook good cases when poor examples were included.
Area of Science:
- Medical Informatics
- Artificial Intelligence in Medicine
Background:
- Expert systems are used for diagnosing thyroid illness.
- The ID3 algorithm is a method for inductive learning.
Purpose of the Study:
- To evaluate the ID3 algorithm's learning capability for thyroid illness diagnosis.
- To assess the impact of data quality and size on ID3 performance.
Main Methods:
- The ID3 algorithm was trained using preclassified patient data for thyroid illness.
- Two datasets were used: unselected (with missing data) and selected (without missing data).
- Classification rules were derived from an existing expert system for thyroid function diagnosis.
Main Results:
- Increased learning cases improved the ID3 algorithm's accuracy up to a plateau.
- The selected learning material (without missing data) yielded better results than the unselected material.
- A weakness was identified: ID3 may fail to learn from good examples if poor examples are present in the dataset.
Conclusions:
- The ID3 algorithm's performance in thyroid illness diagnosis is sensitive to data completeness.
- Data preprocessing and selection are crucial for optimizing ID3's effectiveness.
- The algorithm's limitation in handling noisy or incomplete data requires consideration in practical applications.