deletions | additions       diff --git a/Modeling_approach_used_regularized_linear__.md b/Modeling_approach_used_regularized_linear__.md  index 7568d40..e6ff570 100644  --- a/Modeling_approach_used_regularized_linear__.md  +++ b/Modeling_approach_used_regularized_linear__.md 
...
# Modeling approach  Our main challenge in modeling these data was to strike a balance between predictive accuracy and interpretability. On the one hand, we wanted   to make sure that our model was not overfitting the data and made good predictions, and on the other hand the question of this project required  us to use a method that allowed us to draw inferences about the impact of specific predictors.   To make sure results were easy to interpret, we chose linear regression. To make sure our model does not overfit, we regularized it using L1 regularization (proportion=.9) and L2 regularization  (proportion=.1). This elastic net approach gave us the benefit of sparsity of the solution (due to L1 regularization), while stabilizing our estimates from correlated predictors via a small amount of L2 regularization. We chose the total amount of regularization using 10-fold cross-validation within the training set and chose the value that yielded the lowest average error.  - used regularized linear regression for ease of interpretability  - combined L1 regularization (proportion=.9) and L2 regularization