Data Preprocessing and Homogeneity: The Influence on Robustness and Modeling by PLS Via NIR of Fish Burgers

Fish burgers as new products require their shelf life investigated. Sensory results usually do not follow a homogeneous profile, as it measures human perception. Once the sensory and physicochemical monitoring of the shelf life takes time and considerable investment, the Near Infrared spectroscopy comes as a fast instrumental technique, which can access multiple parameters from the sample at the same time. In order to replace traditional methods improving mathematical modeling, the objective of this study is the estimation of the data preprocessing and homogeneity (Kolmogorov-Smirnov) influence in the quality parameters of Partial Least Squares modeling. Calibration and validation models were evaluated by means of correlation coefficient, Rank, robustness and Residual Prediction Deviation. All the preprocessing available on the software Opus Lab[R] were tested and compared. 72 readings/8 samples of refrigerated grass carp burgers originated the data regarding its water activity, rancid taste, pH and reactive substances of thiobarbituric acid results. The preprocessing methods accessible were Standard Normal Variate, Multiplicative Scatter Correction, 2nd derivative, 1st derivative, Straight Line Subtraction and Min/Max. Each chosen preprocessing generated a model with different parameters. The homogeneity of data proved to have a direct influence on the robustness, confirming the challenge to fit sensory results in Partial Least Squares prediction models. New possibilities to investigate meat products were shown. Keywords: grass carp; hamburgers; NIR; PLS; predictive models; statistics