Prediction of 5-Year Survival with Data Mining Algorithms

Area of Science:

• Oncology
• Data Science
• Machine Learning

Background:

• Accurate survival time prediction is crucial for colon cancer treatment and follow-up.
• Predicting cancer outcomes using clinical data remains a significant challenge.

Purpose of the Study:

• To evaluate ten data mining algorithms for predicting 5-year survival in colon cancer patients.
• To compare the predictive accuracy of machine learning algorithms against physician estimations.

Main Methods:

• Utilized a nationwide German colon cancer dataset from the Robert Koch Institute.
• Applied ten distinct data mining algorithms (e.g., Random Forest, Support Vector Machine) to predict 5-year survival.
• Compared algorithm performance against survival time predictions made by physicians using the same clinical attributes.

Main Results:

• Machine learning algorithms achieved an average accuracy of 67.7% in predicting 5-year survival.
• Physicians' average accuracy in predicting survival time was 59%.

Conclusions:

• Data mining algorithms demonstrate superior accuracy compared to physician predictions for colon cancer survival.
• Machine learning offers a promising tool to enhance the accuracy of prognostic assessments in oncology.