The best approach to preserve the scientific accuracy of the data is to show it as obtained in the instrument. Baseline correction is only necessary when you are performing refinement of the XRD data to extract the lattice parameters and bond angles.

However, especially for polycrystalline thin film samples, where the XRD peak intensity is less (around 1000 cps typically, where bulk powders give a count of 105 or 106 cps), the scattering background becomes prominent and the data does not "look good".

Given that the background does not signify anything important, it is totally acceptable to do a background correction. But it should be indicated in the figure caption. In certain cases, for example, if your sample contains iron, copper k-alpha radiation will give rise to some fluorescence. The shape of the background can also indicate the amorphous phases present in the sample. There are many such examples. In such cases, it will be a good practice to keep the background and explain what it signifies.

In short, understand the origin of the background in your pattern. If it has any relevance in the story of your paper, keep that. If not, it is totally up to you to do a background correction or present the data as is.

"First, thank you very much. Second, my issue when making modifications is the appearance of negative numbers, which for me is a mistake. I want to make adjustments while preserving the values — at the very least, I want to prevent negative values from appearing.

It looks like it is an issue if your choice of background shape. I see many large negative values at the beginning. What shape did you use?

There are many ways of modeling the background. It seems you just need a better one.

Yes, my problem is negative values. I used a linear shape. Please tell me what I can do in this case?

In your screen shot, is A your x values, i.e., two-theta. It looks like C might be your original data. I am not clear on what B is. Is it a second dataset?

I am surprised by the shape of the background in C. I am not used to the signal falling off at lower angles. I am used to it climbing. Could there be something wrong with your geometry, like the sample being set too high? If so, that will lead to other problems with the pattern and its interpretation.

A linear fit is the most basic choice and will not be satisfactory your sample. Piecing two linear regions might give a decent match for your C pattern. A sloped linear model might be a decent fit up to your first large peak. Then a nearly flat background might be a better fit from there to higher angles.

A polynomial or other fit would be better, but you would first need to isolate the background regions to fit. Otherwise, your trend line will try to match the peaks as well as the background.

I have added a screen shot of a pattern from concrete. I only worked with the region up to 25 degrees. The darker line is the background region that I chose to fit. You can see that I left some minor peaks in around 12, 20 and 22.5 degrees that I should have removed. The lighter line is the full pattern.

A 5th order polynomial was a good fit for the background in that area. I think I might need even higher order if I tried to fit the background out to 70 degrees. (Of course, I would need to remove the peaks first, and that is time consuming.)

I also tried fitting a 5th order polynomial to the full dataset. You can see that it is far above the background. The peaks _must_ be removed first.

I will also note that it is usually better to work within software expressly developed for the purpose, e.g., XRD. You can try to use general purpose tools, but it is a lot of reinventing the wheel. The XRD software should have tools for background removal and peak detection and integration. It might also allow for sample height correction.