This work uses data aggregated from a number of sources including the American Community Survey census.gov, clinicaltrials.gov, and cancer.gov 🏥. The dataset contains 3048 datapoints and 33 variables.
🚩 In this project we build and compared several models to predict the "TARGET_deathRate". We tested ordinary least square (OLS) regression, CART and Random Forest algorithms. We also used VSURF for variable selection.
Models were compared using the root mean squared error (RMSE):
| Model | RMSE |
|---|---|
| OLS full dataset (numericals) | 19.259 |
| OLS without outliers | 15.174 |
| Step default parameters | 15.173 |
| Step forward or backward | 15.165 |
| LASSO | 15.248 |
| OLS on LASSO selected features | 15.254 |
| CART final tree | 17.598 |
| CART + VSURF final tree | 17.861 |
| Random Forest | 14.738 |
⚪ We condlude that the best model to predict "TARGET_deathRate" is Random Forest followed by step method forward or backward.