Author |
Message |
suranjan panigrahi (Panispisani)
| Posted on Thursday, September 26, 2002 - 6:58 pm: | |
Please share your thoughts on the following: For prediction of protein content of wheat or grain, what is the average and maximum accurcay that can be obtained by commercially available NIR (lab-scale) instruments.???? thanks suranjan panigrahi |
dardenne
| Posted on Friday, September 27, 2002 - 5:00 am: | |
We got these results for protein in wheat: SECV of 0.26 with 2000 samples (1100-2500nm) on whole wheat grain SECV of 0.22 with 3000 samples (1100-2500nm) on ground wheat SECV of 0.13 with 1000 samples (1100-2500nm) on white flour The acceptable limit would be <0.30 for whole grain. Pierre |
suranjan panigrahi (Panispisani)
| Posted on Friday, September 27, 2002 - 6:02 am: | |
Pierre: Thanks. Would you let me know what is the equation for SECV you used. Also, are you aware of the accuracy of the instrument (lab scale). IS SECV or SEP is the only sufficient measure to express the performance of the equipment/the model. thanks suranjan |
dardenne
| Posted on Friday, September 27, 2002 - 6:47 am: | |
Suranjan, SECV is the root mean square error of cross validation. SEP obtained on an independent set of few samples is generally used. But why trust the statistics from fewer samples than the calibration file? With thousands of samples a 2 segment cross validation is enough and when SEC and SCEV are very closed then I begin to have confidence in the figures. I prefer SECV on 2000 samples than SEP on 20. Pierre |