Related Experiment Videos
Predicting breast cancer survivability: a comparison of three data mining methods
Dursun Delen1, Glenn Walker, Amit Kadam
1Department of Management Science and Information Systems, Oklahoma State University, 700 North Greenwood Venue, Tulsa, OK 74106, USA. delen@okstate.edu
Artificial Intelligence in Medicine
|May 17, 2005
Summary
Decision trees offer the highest accuracy (93.6%) for predicting breast cancer survivability, outperforming artificial neural networks and logistic regression. This study utilized advanced data mining techniques on a large dataset for robust survivability predictions.
Area of Science:
- Biomedical Informatics
- Data Mining
- Oncology
Background:
- Breast cancer survivability prediction is a complex challenge.
- Advancements in technology have enabled collection of large, high-quality datasets.
- Improved analytical methods facilitate effective data processing.
Purpose of the Study:
- To develop and compare prediction models for breast cancer survivability.
- To leverage technological advancements in data collection and analysis.
- To assess the efficacy of different data mining algorithms and statistical methods.
Main Methods:
- Utilized two data mining algorithms: artificial neural networks and decision trees.
- Employed logistic regression, a common statistical method.
- Developed models using a large dataset (>200,000 cases).
- Applied 10-fold cross-validation for unbiased performance estimation.
Main Results:
- Decision tree (C5) achieved the highest accuracy (93.6%) on the holdout sample.
- Artificial neural networks yielded 91.2% accuracy.
- Logistic regression models showed 89.2% accuracy.
- Decision tree performance surpassed previously reported literature accuracies.
Conclusions:
- Comparative analysis provided insights into the relative predictive power of data mining methods for breast cancer survivability.
- Sensitivity analysis on neural networks identified prioritized prognostic factors.
- Decision trees demonstrate superior performance for breast cancer survivability prediction in this study.