# Data best-fit lines: Excessive precision is harmful

The most legendary mathematical tricks, the worst obstacles in the history of physics and all kinds of formulas in which it is difficult for anyone to see the sleeping meaning: these are the inhabitants of Freistetter’s world of formulas.
All episodes of his weekly column, which appears every Sunday, can be viewed here.

If I were to actually analyze my temperature data in the manner described above, I could actually find a function that perfectly describes that dataset – but at most I would be able to make statements about the temperature measured by my weather station in a specific situation (my balcony) ( and only during the period covered by the data). No general conclusions can be drawn from this.

The problem we are dealing with here is called “fitting out” or “fitting out”. polynomial of degree n as a principle n coefficient + 1 so there are several parameters I can modify until the formula ends up matching my data exactly. But precisely because everything is so perfectly formatted, I can be sure that the formula can no longer be used for any other dataset. In addition, it would be very surprising if two weather stations could measure exactly the same data, even if they were right next to each other.

### We don’t need an AI that thinks it’s smarter than it is

If you want to find public connections, you should try to manage with as few freely selectable parameters as possible. In addition, the true law of nature can never reproduce true measurement points and observational data, because all measurements are inevitably subject to errors. Thus, perfect agreement with the interpolation polynomial is the exact opposite of what we are looking for. Rather than finding the underlying relationships between the data, the formula was precisely adapted to the wrong measured values ​​- thus these errors were incorporated into the model.