Page 26 - IJAMD-1-1

International Journal of AI
for Material and Design ML in 3D bioprinting of cultivated meat

Table 5. Summary of published works on meat adulteration detection
Foods ML method Analytical Sample Details Performance References
technique size
Raw beef Unsupervised: FTIR, 55 Inputs: i. PCA: 86
PCA Physicochemical • Data fusion of ATR‑FTIR First two PCs accounted for 79.93% of
Supervised: properties spectral data in the range of the variance.
PLS-DA 4000 – 525 cm and five No clear discrimination using spectral
-1
physicochemical properties of data
protein, ash, sodium, chloride, ii. PLS-DA:
and phosphate content 91% accuracy using three LVs from
Outputs: low-level data fusion
Identification of adulterated samples
Chopped Supervised: E-nose 21 Inputs: i. SVM: 87
beef SVM, Logistic • E‑nose data from nine gas 94.57% accuracy
Regression, sensors ii. Logistic Regression:
Decision Tree, Outputs: 93.71% accuracy
MLP, Ensemble • Classification into seven bins of iii. Decision Tree:
Voting adulterant percentage 91.14% accuracy
SVM Model details: iv. MLP:
• Radial basis function kernel 92.85% accuracy
Logistic Regression Model details: v. Ensemble Voting:
• LBFGS solver 95.71% accuracy
Decision Tree Model details:
• Gini criterion
MLP Model details:
• One hidden layer with ten nodes
• Sigmoidal and tansig functions
Minced Unsupervised: UV-Vis, FT-NIR, 242 Inputs: i. PCA: 88
Beef PCA FT-MIR • FT‑NIR spectral data in the Difficult to distinguish samples with
Supervised: range of 10614 – 3749 cm-1 under 20% turkey adulteration
LDA, PLSR • FT‑MIR spectral data in the Pure turkey was easily distinguished
range of 3701 – 2642 cm-1 and ii. LDA:
2295 – 1008 cm-1 54.6% accuracy using UV-Vis data
• UV‑Vis spectral data in the 71.2% accuracy using NIR data
range of 220 – 700 nm 65.2% accuracy using MIR data
• Data fusion of spectroscopic data iii. PLSR:
Outputs: R = 0.81
2
• LDA: Adulterant concentration RMSEP = 8.61 using UV-Vis data
classification into five bins R = 0.92
2
• PLSR: Adulterant concentration RMSEP = 5.79 using NIR data
estimate R = 0.91
2
RMSEP = 6.19 using MIR data
R = 0.95
2
RMSEP = 5.33 using fused data
Minced Supervised: Compositional 135 Inputs: i. PLSR: 89
beef PLS, SIMCA analysis, FTIR • PLSR: FTIR spectral data in the R > 0.99
2
range of 4000 – 650 cm-1 RMSE = 0.45 for horse meat
2
• SIMCA: FTIR spectral data at R > 0.99
optimal range of 1800 – 900 cm -1 RMSE = 1.39 for soy protein
2
Outputs: R > 0.99
• PLS: Adulterant concentration RMSE = 1.00 for fat beef trimmings
estimate ii. SIMCA:
• SIMCA: Identification of 100% accuracy in recognizing and
adulterant rejecting all adulterants
Beef, Supervised: PLSR FTIR 7 Inputs: R > 0.99 90
2
Meatball • FTIR spectral data at selected RMSEP = 0.71
fingerprint in the range of
1200 – 1000 cm , from fat
-1
extracted using hexane
(Cont’d...)
Volume 1 Issue 1 (2024) 20 https://doi.org/10.36922/ijamd.2279