What is your take on feature selection in machine learning? Why is it so important to consider?

Feature selection is important because bad or random features can be damaging to an ML model and it might actually fit on the noise (overfit). Using only the relevant feature should improve the efficiency and accuracy.

feature selection is an affective way to reduce the complexity of the model, when there is not enough data or we have missing data for some features, moreover, it could help to reduce the bias in the model for some features like gender or race.  There are two ways for doing feature selection: automatic feature selection using techniques like L1 regularization, and the other way is to manually analyze the affect of each feature on the prediction performance. The later is very effective when using traditional feature engineering, however, with representation learning and high dimensional features, it can be more expensive. 

The manual analyziz involves techniques like leave one out and see the affect of the feature on the performance, this could be very expensive for Deep neural networks.