To consider various evaluation features in optimizing the bread formulation while reducing feature redundancy, this paper modifies the principal component analysis (PCA) and introduces a PCA-based unsupervised feature selection (PCAFS) method. A new similarity measure is defined to select features that align with the dominant principal components of feature vectors. The efficiency of the proposed PCAFS is demonstrated through comparative evaluations to find an appropriate subset of features from various UCI benchmark datasets for clustering tasks. To improve the technological properties of 15 % wheat bran enriched bread, first, wheat bran is pretreated. Then, improving agents such as Xylanase, sodium stearoyl-2-lactylate (SSL), and Sorbitol are integrated into the bread’s dough formula. Next, bread image features, including color, texture, and morphological features are extracted. Last, the proposed PCAFS method is applied to select a subset of features for optimizing the bread formula by the response surface method. The experimental results reveal that the optimal sample includes 0.638 % SSL, 0.04 % Xylanase, and 4.784 % Sorbitol. In comparison to two control samples, sensory evaluations and analysis of bread image features demonstrate an improvement in the optimal sample\\\\\\\\\\\\\\\'s textural characteristics. Furthermore, thermal analysis, water activity measurements, and SEM results reveal a reduction in the bread\\\\\\\\\\\\\\\'s retrogradation.

Keywords