Doubts about using regression (Matlab)

I have a data set (Col_A and Col_B) and I need to find the value of Col_B when Col_A is 0.

Col_A Col_B

983 235.2

960 237.1

945 225.9

I've tried several commands, such as fit, fitlm, polyfit, regress but the results are not as expected.

When I use Excel to find the Trendline and find the value if I create the "Insert Scatter (X, Y) or Bubble Chart" graph, I find the values close to what I find in Matlab (which does not correspond with what search) when I create the graph using the "Insert Line or Area Chart" the value found by the trend line is within the expected.

However, I cannot find this value through Matlab.

Codes I've used

fit(Col_A,Col_B,'poly2')

fitlm(Col_A,Col_B,'poly2')

regress(Col_A,Col_B)

polyfit(Col_A,Col_B,2)

polyfit(Col_A,Col_B)

[Rik]

Your data is very far from what you're looking for. Any trend line you will fit to only 3 points will be close to meaningless if you don't know the underlying function and only need to fit the parameters.

Can you confirm you want to fit a second order polynomial?

[Juliano Cardoso]

Some samples have more than 3 data, the ideal would be to use the same adjustment for all samples, so I was trying with a polynomial, but more importantly, the value found should be close to the expected value

[Rik]

Your data range is 945-983. How will you prevent garbage outcomes if you try to extrapolate to 0? You can indeed use a polynomial, but you need to make sure you know enough about the system you're modeling to decide if that makes sense. You could also fit a sine wave to your data.

So again: are you very sure you need a second order polynomial specifically?

[Juliano Cardoso]

I just need an approximate and plausible value for for Col_B when Col_A = 0, and I used a second order polynomial just because the curves are not linear, but I could use a first order polynomial without any major problems. I need this value only as an initial condition for other calculations and therefore I need an approximate value only.

[Cris LaPierre]

Perhaps you are over complicating it. If a linear model would work as well as a 2nd order polynomial for your needs (which would result in very different values), what about just hard coding a generic value?

[Juliano Cardoso]

Because in practice this value changes every moment (but they are all in a range of values) and if I consider only one value the final results would practically not vary (I already tried this) and I would not like to assign random values.

[Rik]

If you want a different value every time you could also consider using the mean of your data.

[Juliano Cardoso]

I think using mean will be the simplest solution