A Machine Learning Specialist is working on a linear regression model and notices the model is overfitting. The specialist applies an L1 regularization parameter and runs the model again…

Increasing the L1 regularization parameter would further strengthen the penalty on coefficients, likely keeping all weights at zero or making the model even simpler and less effective.

If all features have zero weights after applying L1 regularization, it means the regularization parameter is too high, excessively penalizing the coefficients. Decreasing the L1 regularization parameter will reduce this penalty, allowing some feature weights to become non-zero and for the model to learn.

While L2 regularization is another technique for overfitting, the immediate problem is excessive L1 regularization, which caused all weights to go to zero. Changing to L2 or increasing its parameter without addressing the L1 issue directly isn't the most direct solution.

Decreasing the L2 regularization parameter would reduce its penalty, potentially making the model more complex, which is counterproductive when addressing an overfitting issue (even if the current L1 over-regularization caused underfitting).