On the quality of neural net classifiers

Area of Science:

• Computer Science
• Machine Learning
• Medical Informatics

Background:

• Neural network classifiers are increasingly used in medical diagnostics.
• Assessing the overall quality of these classifiers requires metrics beyond simple accuracy.
• Understanding misclassification patterns is crucial for clinical utility.

Purpose of the Study:

• To introduce novel concepts and metrics for evaluating neural network classifier quality.
• To provide tools for both designers and users to assess classifier utility.
• To address specific aspects of classifier misclassifications.

Main Methods:

• Development of new quality concepts and associated metrics for neural network classifiers.
• Application of these metrics to compare multiple neural network classifiers.
• Evaluation within the specific domain of thyroid disorder diagnosis.

Main Results:

• Introduction of a comprehensive set of metrics for neural network classifier quality assessment.
• Demonstration of the utility of these metrics in identifying superior classifiers.
• Highlighting the importance of considering misclassification characteristics.

Conclusions:

• The proposed metrics offer a robust framework for assessing neural network classifier quality.
• These metrics are applicable and beneficial for practical classifier selection in healthcare.
• The study provides valuable insights for improving the reliability of AI in medical diagnostics.