Evaluating learning models with transitions of human interests based on objective rule evaluation indices

Area of Science:

• Data Mining
• Human-Computer Interaction
• Medical Informatics

Background:

• Post-processing of data mining results is crucial for success.
• Evaluating numerous rules from noisy, large datasets is challenging for human experts.
• Existing methods lack objective criteria for rule evaluation.

Purpose of the Study:

• To develop a method supporting data mining evaluation using human-system interaction.
• To introduce objective rule evaluation indices for mathematical assessment of if-then rules.
• To analyze the performance of rule evaluation models using actual data mining results.

Main Methods:

• Designed a method utilizing objective rule evaluation indices.
• Calculated indices to mathematically evaluate each if-then rule.
• Applied five representative learning algorithms to construct rule evaluation models.
• Used a chronic hepatitis dataset for evaluation.

Main Results:

• Developed and validated objective indices for rule evaluation.
• Demonstrated the effectiveness of the proposed method in supporting expert evaluation.
• Established rule evaluation models based on actual data mining outcomes.
• Observed relationships between expert subjective criteria and model performance.

Conclusions:

• The proposed method effectively supports the data mining evaluation process.
• Objective indices provide a mathematical basis for rule assessment, overcoming expert limitations.
• The study highlights the interplay between computational evaluation and expert judgment in data mining.