A linear regression model was overfitting, so you applied L1 regularization. After training, every feature weight became zero. What is the best next step to improve the model?

If all feature weights became zero after applying L1 regularization, it indicates that the regularization strength was too high. L1 regularization (Lasso) encourages sparsity by driving less important feature weights to zero. When all weights are zero, the model is underfitting severely, essentially predicting the mean of the target variable without considering any features. Therefore, reducing the L1 regularization strength will allow the model to assign non-zero weights to relevant features, improving its ability to learn from the data. Increasing L1 regularization would exacerbate the problem. Adding L2 regularization (Ridge) would not address the fundamental issue of excessive L1 penalty causing all weights to vanish.