Thesis
Predicting sensory attributes found in a model wine using singular value decomposition and support vector machines
Washington State University
Master of Science (MS), Washington State University
2016
Handle:
https://hdl.handle.net/2376/101297
Abstract
Advanced mathematical modeling can be applied to give insight into dynamic systems, including wine. Two of these mathematical models include Singular Value Decomposition and Support Vector Machines. This thesis used both SVD and SVR to data mine, examine overall sensory panel performance, and identify outliers from a previously conducted study on model red wines. In this previous study, twelve trained panelists rated the intensity of 20 different sensory attributes in model red wines varying in ethanol, tannin and fructose concentrations. These evaluation scores were analyzed using standard statistical methods including analysis of variance (ANOVA) and Principal Component Analysis (PCA). Our study went beyond these traditional methods and applied advanced mathematical modeling to these intensity attribute ratings thus predicting panelist perception based on the composition of the sample. Therefore, taking the data from the previous study and the known composition of the model wine, the thesis applied additional statistical and mathematical methods to create predictive models describing sensory perceptions. Leaving out the panelists while re-running the model over multiple iterations with every combination of panelists removed, trained panelist who were considered "outliers" were identified using a "leave out then re-run" scenario. Using SVD and SVR, the sensory response of the wines could be predicted from the chemical composition of the wine, with prediction rates of prediction: r = 0.9077 (first order SVD), 0.9170 (second order), and 0.9245 (third order), respectively. The first order SVD corresponds to the optimal linear model. The SVR method did not perform as well (r = 0.9198) as the higher order approximations using SVD although, this was likely due to incomplete model optimization. Using the methods developed by this work, several future applications for datamining were contemplated, including smartphone app development for consumer preference, models to guide winemaker blending decisions, robotics and sensor array development, and identification of outliers who may represent a niche market.
Metrics
11 File views/ downloads
17 Record Views
Details
- Title
- Predicting sensory attributes found in a model wine using singular value decomposition and support vector machines
- Creators
- Daniel Archer Dycus
- Contributors
- Carolyn Felicity Ross (Degree Supervisor)
- Awarding Institution
- Washington State University
- Academic Unit
- Food Science, School of
- Theses and Dissertations
- Master of Science (MS), Washington State University
- Publisher
- Washington State University; [Pullman, Washington] :
- Identifiers
- 99900525117901842
- Language
- English
- Resource Type
- Thesis