Normalization
Normalization is a data preprocessing technique that transforms numerical features to a common scale. It helps prevent features with larger numerical ranges from disproportionately influencing a machine learning model.
What is Normalization?
Normalization typically rescales values to a defined range, such as 0 to 1, while preserving their relative relationships. The technique is often applied before model training, particularly when input features have substantially different scales. It is distinct from standardization, which typically transforms data based on its mean and standard deviation.
Why is Normalization Important?
Normalization can improve the stability and efficiency of model training by putting features on comparable scales. It is particularly useful for algorithms that are sensitive to feature magnitude or rely on distance calculations, such as neural networks and k-nearest neighbors.
Common use cases
Normalization is commonly used in neural networks, image processing, k-nearest neighbors, clustering, gradient-based optimization, and other machine learning applications.