Premium
Prediction and qualitative analysis of sensory perceptions over temporal vectors using combination of artificial neural networks and fuzzy logic: Validation on Indian cheese (paneer)
Author(s) -
Chaturvedi Kartikey,
Khubber Sucheta,
Singha Siddhartha,
Goel Himanshu,
Barba Francisco J.,
Das Kalyan
Publication year - 2020
Publication title -
journal of food processing and preservation
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.511
H-Index - 48
eISSN - 1745-4549
pISSN - 0145-8892
DOI - 10.1111/jfpp.14955
Subject(s) - sensory system , artificial neural network , backpropagation , artificial intelligence , food science , mathematics , pattern recognition (psychology) , ranking (information retrieval) , fuzzy logic , machine learning , computer science , biology , neuroscience
The present study investigated the feedforward‐backpropagation artificial neural network (ANN) architectures to predict the sensory scores at different moisture levels (40%–50%) of paneer. Paneer was produced aseptically and packed in low density polyethylene (LDPE) bags, in a semi‐organized commercial dairy plant. Samples were evaluated at regular intervals (8 days) for biochemical content and microbial counts, while subjected every day for sensory evaluation. Three layered (input‐hidden‐output) ANN was able to produce similar sensory responses using biochemical and microbiological data, for both single (best combination 10‐7‐1; R 2 ≈ 0.99) and conjugated (10‐9‐4; R 2 ≈ 0.97) parameters for an extent of 25 sensory output nodes (10‐35‐25; R 2 ≈ 0.90). The comparison of ANN and predicted sensory scores using linear regression model (with R 2 = 0.99) suggests that paneer at 42%–44% and 47%–49% moisture was best for consumption as fresh and for storage. The developed method for predicting and analyzing sensory data of produced paneer opens new possibilities for improving food product's likeness. Practical applications Current fuzzy models probe quality ranking of complex sensory responses as discrete values, normally onset of manufacturing. However, products are stored for a longer period before reaching to the consumers. The results of the present findings suggest applications of machine learning, such as artificial neural network, for predicting sensory performance of the product throughout the shelf using few microbiological and biochemical parameters. The developed fuzzy scale for approaching qualitative performance of the product in terms of sensory likeness would enable manufacturers to account changes throughout the shelf, providing new solutions for problems in sensory data analysis. Henceforth, opening possibilities of combining fuzzy logic and machine learning to obtain robust and conclusive data analysis for quality comparison for stored product, while enabling ANN to predict them.