Abstract
This study involved a comprehensive examination of sensory attributes in dry-cured Bísaro loins, including odor, androsterone, scatol, lean color, fat color, hardness, juiciness, chewiness, flavor intensity and flavor persistence. An analysis of 40 samples revealed a wide variation in these attributes, ensuring a robust margin for multivariate calibration purposes. The respective near-infrared (NIR) spectra unveiled distinct peaks associated with significant components, such as proteins, lipids and water. Support vector regression (SVR) models were methodically calibrated for all sensory attributes, with optimal results using multiplicative scattering correction pre-treatment, MinMax normalization and the radial base kernel (non-linear SVR model). This process involved partitioning the data into calibration (67%) and prediction (33%) subsets using the SPXY algorithm. The model parameters were optimized via a hybrid algorithm based on particle swarm optimization (PSO) to effectively minimize the root-mean-square error (RMSECV) derived from five-fold cross-validation and ensure the attainment of optimal model performance and predictive accuracy. The predictive models exhibited acceptable results, characterized by R-squared values close to 1 (0.9616-0.9955) and low RMSE values (0.0400-0.1031). The prediction set's relative standard deviation (RSD) remained under 5%. Comparisons with prior research revealed significant improvements in prediction accuracy, particularly when considering attributes like pig meat aroma, hardness, fat color and flavor intensity. This research underscores the potential of advanced analytical techniques to improve the precision of sensory evaluations in food quality assessment. Such advancements have the potential to benefit both the research community and the meat industry by closely aligning their practices with consumer preferences and expectations.