Related Experiment Videos
Canonical correlation analysis for data reduction in data mining applied to predictive models for breast cancer
Amir Reza Razavi1, Hans Gill, Hans Ahlfeldt
1Department of Biomedical Engineering, Division of Medical Informatics, Linköping University, Sweden.amirreza.razavi@imt.liu.se
Studies in Health Technology and Informatics
|September 15, 2005
Summary
This study introduces a new preprocessing method to identify key predictors for breast cancer recurrence. The method improves predictive model accuracy by reducing data complexity before analysis.
Area of Science:
- Biostatistics
- Medical Informatics
- Oncology
Background:
- Predicting breast cancer recurrence is crucial for patient management.
- High-dimensional datasets pose challenges for developing accurate predictive models.
- Effective variable selection is essential for data mining in medical research.
Purpose of the Study:
- To present a novel preprocessing method for selecting important variables in breast cancer datasets.
- To enhance the accuracy of predictive models for breast cancer recurrence.
- To reduce the complexity of large datasets for more efficient analysis.
Main Methods:
- Data from 5787 female patients were integrated from breast cancer, tumor marker, and cause of death registers.
- Logical rules were applied for data cleaning, followed by domain expert selection of predictors.
- Canonical Correlation Analysis (CCA) was used for dimension reduction, and Artificial Neural Networks (ANN) for predictive modeling, validated by ten-fold cross-validation.
Main Results:
- The proposed preprocessing method effectively identified important predictors for breast cancer recurrence.
- The application of CCA and ANN resulted in improved prediction accuracy.
- A significant reduction in mean absolute error was observed, indicating enhanced model performance.
Conclusions:
- The developed preprocessing technique is effective in variable selection for breast cancer recurrence prediction.
- Integrating data mining techniques like CCA and ANN can significantly improve predictive model performance.
- This approach offers a valuable tool for medical data analysis and knowledge discovery in oncology.