we use Surveillance, Epidemiology, and End Results (SEER) Incidence Data to predict the chance of survival given cancer diagnosis information obtained when the patient is diagnosed with cancer. We find answers to two questions:
Question 1: Given the demographic and diagnosis information, can we predict if the patient will pass away due to complications caused by the diagnosed cancer?
Question 2: For patients who eventually pass away due to cancer, can we predict the number of months they expect to survive?
The project is formulated in the Jupyter notebook Cancer_survival_prediction.ipynb. It has different sections to load, clean, and process data. The jupyter notebook is self-contained and includes detailed information about the project.
The data in this project comes from Incidence - SEER Research Data, 18 Registries, Nov 2020 Sub (2000–2018). It is a public dataset that collects cancer incidence data from several registries across the U.S. The data (.csv files) are not included in the repository because one needs to request access to the data. If you need to access the dataset, you must submit a request. Also, the .csv files are very large (over 6GB).
This project requires Python 3.x and the following Python libraries:
- Machine Learning Libraries: NumPy, Pandas, Sciki-Learn, tensorflow, xgboost
Please feel free to use the code here as you would like.