The process of rescaling features to a common range, typically [0,1] or z-scores. In forex ML, prices and indicators can have widely different scales (e.g. pip values vs. percentage). Normalization (min-max scaling) or standardization (zero mean, unit variance) ensures each input contributes proportionately. This is especially important for algorithms like k-NN or neural networks, and speeds up optimization (gradient descent converges faster with normalized features). For example, using scikit-learn’s StandardScaler before training is common practice.

Back to Robots Glossary Index