Research on parameter selection and optimization of C4.5 algorithm based on algorithm applicability knowledge base.

Area of Science:

• Machine Learning
• Data Mining
• Medical Informatics

Background:

• The C4.5 decision tree algorithm offers high prediction accuracy and interpretability for medical datasets.
• Hyperparameter tuning is crucial for optimizing C4.5 model performance but can be time-consuming.
• Efficient hyperparameter selection is needed for rapid and accurate model development.

Purpose of the Study:

• To optimize the selection of hyperparameters for the C4.5 algorithm.
• To develop a method for fast and accurate optimization of C4.5 algorithm models.
• To reduce unnecessary hyperparameter tuning for improved efficiency.

Main Methods:

• Constructed decision tree models with varying hyperparameter values.
• Evaluated model performance using accuracy, AUC, and F1-measure.
• Integrated performance data with dataset metadata to build a hyperparameter optimization meta-database.
• Utilized modeling learning to recommend hyperparameter ranges for different datasets.

Main Results:

• For over 65% of datasets, hyperparameter M tuning is unnecessary, saving significant time.
• The developed hyperparameter optimization model achieved an accuracy rate exceeding 80%.
• The recommended hyperparameter values were validated, proving their feasibility.

Conclusions:

• The study provides a feasible method for fast tuning and optimization of C4.5 algorithm parameters.
• The findings reduce the need for extensive hyperparameter tuning, enhancing efficiency.
• This research offers a valuable basis for optimizing C4.5 models in medical applications.