Sensor-Based Evaluation of Purslane-Enriched Biscuits Using Multivariate Feature Selection and Spectral Analysis

This study presents a sensor-integrated framework for evaluating purslane (Portulaca oleracea L.) stalk flour as a functional ingredient in butter biscuits. A Design of Experiments (DoEs) approach was applied using multisensor probes (electrical conductivity, pH, TDS, ORP) and digital imaging sensors (visible reflectance spectra) for real-time, non-destructive quality assessment. Multivariate analysis with Repeated Relief Feature Selection (RReliefF) and Principal Component Analysis (PCA) reduced 54 initial measurements to 19 informative features, with the first two principal components explaining over 96% of the variance related to flour concentration. Regression modeling combined with linear programming identified an optimal substitution level of 9.62%. Biscuits at this level showed improved texture, enhanced elemental composition (Ca, Mg, Fe, Zn), stable color, and maintained sensory acceptability. The methodology demonstrates a reliable, low-cost sensing and chemometric approach for data-driven, non-destructive quality monitoring and product optimization in food manufacturing.