Min-Max Normalization

Min-max normalization linearly stretches or compresses a value so the smallest value in a feature becomes 0 and the largest becomes 1. It works by measuring how far x is above the minimum, then dividing by the full data range, so the result is the fraction of the interval from minimum to maximum that x occupies. This preserves the original ordering of values and keeps the spacing proportional, just on a standardized scale.

Use this when different numeric features need to be compared on the same scale, such as before distance-based methods like k-nearest neighbors or clustering. It is also useful when you want all inputs to lie in a fixed range, especially [0, 1], for neural networks or visualizations.

• Using the sample's current value as the minimum or maximum instead of the true feature-wide min and max
• Forgetting to subtract the minimum before dividing, which changes the scale incorrectly
• Applying min-max normalization to new test data using test-set min and max instead of the training-set values