Some observations made:

 R, G, B and Greyscale values in the given xlsl file are not consistent. The first problem with the dataset is, blue channel values and red channel values have been interchanged.
Second, the color of the sample zone doesn't match with that of the image (more skewed towards red value). Maybe some man-made error while extracting the data. 
The RGB values in the reference zone are relatable with that of the image, but, again, the values are very inconsistent at the initial set of images (i.e, images for low dilution factor). Based on this observation, it is obvious a model trained with this dataset would not give a good result. However, I tried to visualize using linear regression.