It depends what you mean by best.
If you have n discrete points, you can always make a polynomial curve go though every single point by using order n-1.( Two points - straight line order 1, Three points - parabola order two, Four points - cubic, order three etc.) That is probably not what you want in a real world problem with many points because the higher the order the bumpier the curve (straight line - constant direction, parabola first up then down, cubic up- down - up etc). In other words you need to choose your model based on what is appropriate for the type of problem you are working.
How do you find the best and simplest model given a table of data? I know how to find a specific regression (linear, cubic ect...) but i'm lost when it comes to choosing the best one.
1 answer