Data Science and Machine Learning (Theory and Projects) A to Z - Feature Selection: Embedded Methods

The video tutorial discusses three feature selection methods: filter, wrapper, and embedded. Filter methods are fast but lack model specificity. Wrapper methods are model-specific but time-consuming due to repeated training. Embedded methods combine the advantages of both, being fast and model-specific by training once and using weights to determine feature importance. An example using L1 regularization (lasso regression) illustrates how embedded methods work. The video concludes with a comparison of the methods and a preview of future topics, including implementation in Python.

What is a key disadvantage of wrapper methods in feature selection?

• They are not time-consuming.

• They do not use a machine learning model.

• They require extensive retraining for different subsets.

• They are not model-specific.

How do embedded methods differ from wrapper methods in terms of training?

• Embedded methods are not model-specific.

• Embedded methods train the model only once.

• Embedded methods do not use a machine learning model.

• Embedded methods train the model multiple times.

What role do weights play in embedded methods?

• Weights indicate the importance of features.

• Weights are used to select the machine learning model.

• Weights determine the speed of the model.

• Weights are irrelevant in embedded methods.

What is the primary function of L1 regularization in feature selection?

• To maximize the model's complexity.

• To minimize weights and identify unimportant features.

• To ensure all features are equally important.

• To increase the number of features.

Which feature selection method is known for being fast and not requiring model specificity?

• L1 regularization

• Filter method

• Embedded method

• Wrapper method

What is a common characteristic of both wrapper and embedded methods?

• They both require multiple training sessions.

• They are not model-specific.

• They are both model-specific.

• They do not use machine learning models.

Why might using features selected by embedded methods in a different model be problematic?

• Embedded methods are not model-specific.

• Features are highly specific to the model used in training.

• Embedded methods do not select features.

• Features are universally applicable to all models.