Which assertion about independent variable data types in neural networks is correct?

Which assertion about independent variable data types in neural networks is correct?

• A. Only categorical data can be used
• B. Numeric data type is required for independent variables
• C. Text data types are preferred for modeling
• D. Any data type may be used for independent variables

Answer: (B)

Neural networks operate through arithmetic on inputs, weights, and activations. Everything in those computations must be numeric, because matrix multiplications, additions, and activation functions only accept numbers. If a feature isn’t numeric, you must transform it into a numeric representation before feeding it into the model (for example, one-hot encoding for categories or embeddings for text). That’s why the statement that a numeric data type is required for independent variables is the correct one: it captures the essential requirement that the network’s inputs be numeric, with non-numeric features converted into numbers prior to modeling. The other ideas don’t fit because raw categories or text aren’t used directly by neural nets; they must be encoded into numeric form first, and saying any data type can be used as-is ignores this encoding step. Similarly, focusing on text as the preferred data type is misleading since it also requires numeric representation before the network can process it.

Neural networks operate through arithmetic on inputs, weights, and activations. Everything in those computations must be numeric, because matrix multiplications, additions, and activation functions only accept numbers. If a feature isn’t numeric, you must transform it into a numeric representation before feeding it into the model (for example, one-hot encoding for categories or embeddings for text). That’s why the statement that a numeric data type is required for independent variables is the correct one: it captures the essential requirement that the network’s inputs be numeric, with non-numeric features converted into numbers prior to modeling.

The other ideas don’t fit because raw categories or text aren’t used directly by neural nets; they must be encoded into numeric form first, and saying any data type can be used as-is ignores this encoding step. Similarly, focusing on text as the preferred data type is misleading since it also requires numeric representation before the network can process it.