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

The video tutorial discusses various feature selection methods, focusing on filter methods that do not rely on machine learning models. It explains the low variance criteria, which eliminates features with low variation, and the T score criteria, used for binary classification to maximize class separation. The Chi-squared score, suitable for multiclass problems, tests feature independence from class labels. Advanced criteria like the Hilbert-Schmidt independence criterion are also mentioned. The tutorial highlights the limitations of statistical methods in handling feature redundancy.

What is the primary focus of filter methods in feature selection?

• Combining features to create new ones

• Selecting features based on predefined criteria without a model

• Using a machine learning model to guide feature selection

• Eliminating features based on user preference

Which of the following best describes the low variance criterion?

• A technique that maximizes the difference between class means

• A supervised technique that uses class labels

• An unsupervised technique that eliminates features with low variation

• A method that requires subset generation

What is the significance of a threshold in the low variance criterion?

• It is a fixed value for all datasets

• It is used to normalize feature values

• It sets the significance level for feature selection

• It determines the computational cost

What is the main goal of the T-score criterion in feature selection?

• To eliminate features with low variance

• To test the independence of features

• To minimize the variation within classes

• To maximize the variation between two classes

What is the relationship between T-score and Fisher score?

• Both are unsupervised methods

• Both aim to minimize within-class variation

• Both require subset generation

• Both are used for multiclass problems

Which feature selection method is suitable for multiclass problems?

• Low variance criterion

• T-score criterion

• Chi-squared score

• Wrapper methods

What does the Chi-squared score test in feature selection?

• The variance of features

• The computational cost of feature selection

• The correlation between features

• The independence of a feature from the class label

What is a limitation of statistical-based feature selection methods?

• They are computationally expensive

• They cannot handle feature redundancy

• They require a large dataset

• They are only applicable to binary classification

How do information-theoretical methods improve feature selection?

• By simplifying the feature selection process

• By reducing computational cost

• By using class labels more effectively

• By addressing feature redundancy

Which method is known for transforming features to a higher dimensional space?

• Chi-squared score

• T-score criterion

• Hilbert-Schmidt independence criterion

• Low variance criterion