A Framework for Predictive Modeling of Sensory-Driven Food Quality Systems

Thesis Abstract

This study addresses a critical gap in food quality management by integrating sensory science with predictive modeling to operationalize sensory-driven quality decisions across product development, processing, and consumer evaluation. The central aim is to develop a robust framework for predictive modeling that links sensory attributes to objective quality outcomes, enabling proactive control of product performance in the marketplace. Specific objectives include (i) identifying core sensory descriptors that drive perceived quality across representative food categories (dairy, processed meat, and bakery products), (ii) constructing and validating a hierarchical predictive model that maps sensory profiles to instrumental and consumer-based quality metrics, (iii) evaluating the temporal stability of sensory-driven quality via longitudinal data, and (iv) devising a decision-support tool that integrates model outputs into quality management workflows. The study adopts a multi-method approach underpinned by established theories in sensory science and quality management, notably the Multivariate Statistical Theory of Sensory Data and the Total Quality Management framework, with supplemental grounding in the Expectancy-Disconfirmation theory to explain consumer satisfaction outcomes. A mixed-methods research design combines quantitative data from sensory panels, instrumental analyses, and consumer tests with qualitative insights from expert interviews to refine model structure. The population comprises trained sensory panelists (n=40) for descriptive profiling, instrumental measurements (texture, color, rheology, and volatile compounds) from a dataset of 150 production batches across three product categories, and a consumer cohort (n=600) stratified by age and region for acceptability testing. A stratified random sampling technique yields 50 batches for in-depth modeling and 100 batches for model validation, with consumer testing conducted on a subset of 300 respondents. Data collection instruments include a standardized descriptive analysis protocol, calibrated instrumental sensors (Texture Analyzer, Colorimeter, GC-MS for volatiles), a validated consumer acceptability questionnaire, and semi-structured interviews with quality managers to capture operational constraints. Validity and reliability are addressed through inter-panelist calibration with repeated measures (intra-class correlation >0.85 for key descriptors), instrument calibration traces, and test-retest reliability for questionnaires (Cronbach’s alpha >0.90). Data analysis proceeds in three stages (i) exploratory data analysis and feature selection using Principal Component Analysis and Correlation Network analysis to identify sensory-attribute clusters, (ii) development of a hierarchical mixed-effects predictive model that integrates sensory descriptors (level 1), instrumental metrics (level 2), and consumer acceptability and shelf-life indicators (level 3). Model estimation employs regression-based techniques (LASSO and Elastic Net) to handle multicollinearity, followed by partial least squares structural equation modeling to capture latent relationships. Model performance is evaluated through cross-validation, root mean square error, R-squared, and area under the ROC curve for classification of high versus low-quality outcomes. The framework includes a Bayesian updating component to refine predictions with new batch data. Expected findings indicate that a core subset of sensory attributes (e.g., aroma intensity and persistence, mouthfeel smoothness, and aftertaste balance) exerts disproportionately strong influence on consumer acceptability and shelf-stability predictions, with instrumental proxies explaining a substantial portion of variance in quality outcomes (R2 > 0.70 in validation sets). The study anticipates that integrating sensory data with instrumental measures enhances predictive accuracy and supports early-stage decision-making in formulation and process control. The contribution to knowledge lies in a replicable framework that formalizes the linkage between perceptual quality and objective performance, bridges sensory science with predictive analytics, and provides a scalable decision-support tool for food manufacturers seeking to optimize sensory-driven quality across product portfolios. The main conclusion is that predictive modeling informed by robust sensory profiling can reliably forecast quality trajectories and guide proactive quality assurance. Recommendations include integrating the framework into existing quality management systems, expanding the model to additional product categories, and employing real-time sensory analytics through rapid sensory testing modalities to further shorten development cycles and reduce waste.