Min-Max Scaling
Rescaling a column into a fixed range, usually 0 to 1, by subtracting the minimum and dividing by the span.
Think of It Like This
Like using an image editor to stretch a dim photo so the darkest pixel becomes pure black and the brightest pixel becomes pure white.
. Use it when something downstream genuinely needs bounded input, and prefer standardisation otherwise.
Its weakness is structural rather than incidental: both parameters come from a single row each. One extreme value sets max, so the rest of the data gets squeezed into a sliver near zero. On the column [10, 12, 11, 13, 1000] the four ordinary values land inside the first 0.3% of the range.
Also worth knowing that a value outside the training range maps outside 0 to 1 at serving time, so the guarantee people rely on quietly stops holding on new data.