La normalisation min-max est une technique de prétraitement des données used to scale numerical features to a specific range, usually between 0 and 1. This method transforms the original data points into a normalized scale, making it easier for apprentissage automatique algorithms to process the data effectively. The formula used for min-max normalization is:
Xnorm = (X – Xmin) / (Xmax – Xmin)
Où :
- Xnorm is the valeur normalisée.
- X est la valeur d'origine.
- Xmin is the minimum value in the dataset.
- Xmax est la valeur maximale dans l'ensemble de données.
By applying this transformation, the data is reshaped so that it fits within the desired range, which helps in reducing the effects of outliers and improving convergence during model training. Min-Max normalization is particularly useful for algorithms that are sensitive to the scale of data, such as neural networks and k-plus proches voisins.
However, it’s important to be aware that min-max normalization can be sensitive to outliers since they can significantly affect the minimum and maximum values. Therefore, it may be advisable to use other Techniques de normalisation si l'ensemble de données contient des valeurs extrêmes.