Breast cancer data analysis for survivability studies and prediction

Area of Science:

• Oncology
• Data Science
• Machine Learning

Background:

• Breast cancer is a leading cause of mortality in females globally.
• Accurate survivability prediction remains a significant challenge.
• Current methods rely on statistical or supervised machine learning techniques.

Purpose of the Study:

• To develop a robust data analytical model for breast cancer survivability.
• To improve understanding of survivability with missing data.
• To identify factors influencing patient survival and establish patient cohorts.

Main Methods:

• Utilized unsupervised learning: Self-Organizing Map (SOM) and DBSCAN for patient clustering.
• Employed Multilayer Perceptron (MLP) for survivability analysis on identified cohorts.
• Applied information gain for variable selection on SEER dataset.

Main Results:

• SOM and DBSCAN identified nine distinct patient cohorts with varying survivability.
• Clustering patients improved MLP survivability prediction accuracy.
• Revealed complex factors influencing prediction accuracy and patient outcomes.

Conclusions:

• A novel, data-driven unsupervised learning approach enhances breast cancer survivability prediction.
• Patient segmentation into cohorts provides better insights into subgroup-specific survival.
• This method improves prediction accuracy compared to using raw historical data.