Establishing a Berry Sensory Evaluation Model Based on Machine Learning.

Area of Science:

• Agricultural Science
• Food Science
• Computational Science

Background:

• Rising quality of life increases demand for high-quality fruits like blueberries.
• Sensory evaluation is crucial for blueberry quality but is subjective.
• Objective, data-driven methods are needed to complement traditional sensory analysis.

Purpose of the Study:

• To develop an objective machine learning model for blueberry quality assessment.
• To compare the performance of a particle swarm optimization support vector regression model against other machine learning approaches.
• To reduce human subjectivity in blueberry sensory evaluation.

Main Methods:

• Utilized ten physical and chemical blueberry indices (e.g., catalase, flavonoids, soluble solids) as input features.
• Developed a support vector regression model optimized via particle swarm optimization (PSO-SVM).
• Compared PSO-SVM with convolutional neural networks (CNN) and long short-term memory (LSTM) networks, repeating experiments 20 times.

Main Results:

• The PSO-SVM model achieved lower error metrics: root mean square error (RMSE) of 0.45 and mean absolute error (MAE) of 0.40.
• CNN and LSTM models showed higher errors: RMSE (0.96, 1.22) and MAE (0.78, 0.97), respectively.
• The PSO-SVM model demonstrated superior predictive accuracy, especially with limited sample data.

Conclusions:

• Particle swarm optimization effectively enhances support vector regression for blueberry quality prediction.
• The proposed PSO-SVM model offers a more reliable and objective alternative to traditional sensory evaluation.
• This approach is particularly advantageous in scenarios with limited blueberry sample data.