This is the usual result: near infrared spectra are very good for prediction of total soluble solids in fruits (also water or total dry matter), acids are tricky and texture never works. The articles that claim otherwise have "autocorrelated" samples i.e. for example a single variety, in which the evolution of TSS is correlated to the loss in acidity and texture as maturation processes. I've also seen "two-points correlations" with two contrasted varieties, but it is the same idea: when the TSs is in intself a good predictor of texture or acidity (i.e. strongly correlated to texture or acidity) it looks like NIR works, but it is an artefact. When sample sets are orthogonalized, NIR samples predict TSS from fruits, which means usually sugars or sometimes acidity in highly acidic fruits such as lemons or passion fruit.

Thank you Dr Catherine. But i can see in many related published articles in which the authors have modeled the firmness, TA acidity and SSC with very good statistics results. I am in doubt on my preprocessing of spectral data. Also, unlike the other research, by applying the proposed preprocessing transforms in related publication, I can't reach to acceptable results.