Assessment of Various Machine Learning Models for Peach Maturity Prediction Using Non-Destructive Sensor Data.

Area of Science:

• Agricultural Science
• Machine Learning
• Data Science

Background:

• Peach maturity prediction is crucial for harvest timing and quality control.
• Non-destructive methods are preferred for assessing fruit ripeness without damage.
• Previous studies utilized various machine learning models, but a comparative analysis was lacking.

Purpose of the Study:

• To compare the performance of eight machine learning models for peach maturity prediction.
• To identify the most accurate model for predicting 'Suncrest' peach ripeness using non-destructive data.
• To evaluate the contribution of specific peach traits to maturity prediction through variable subgroup selection.

Main Methods:

• Eight machine learning models were trained on a dataset of 180 'Suncrest' peaches.
• Dimensionality reduction was performed using least absolute shrinkage and selection operator (LASSO) regularization, selecting 8 of 29 input variables.
• Group LASSO regularization was used to identify the contribution of peach ground color measurements.

Main Results:

• The artificial neural network (ANN) model achieved the highest performance with an Area Under the Curve (AUC) of 0.782.
• Linear discriminant analysis (LDA) was the second-best model with an AUC of 0.766.
• Accuracy, F1 score, and kappa confirmed ANN's superior performance in peach maturity prediction.

Conclusions:

• Artificial neural network (ANN) is the most accurate machine learning model for predicting peach maturity based on the evaluated non-destructive dataset.
• Variable subgroup selection, particularly using group LASSO, provides insights into trait importance for maturity prediction.
• The findings offer a robust method for optimizing harvest timing and ensuring peach quality.