Analysis of Avocado Ripening with Spectroscopy Using NIR and Applying Machine Learning Techniques

Ovalle C.; Franco A.; Aching O.S. · Proceedings of the International Conference on Soft Computing and Machine Intelligence, ISCMI (2025) · DOI: 10.1109/iscmi67495.2025.11358467

Resumen

This study evaluates the ripeness status of avocados (Persea americana Mill.) of the 'Hass' and 'Fuerte' varieties by near-infrared (NIR) spectroscopy. 64 fruits were analyzed, obtaining four spectra per sample. Subsequently, the dry matter content (DMC) was determined in order to correlate it with the state of maturity. The spectra were preprocessed using the SNV, MSC, the first Savitzky-Golay derivative and principal component analysis (PCA) methods, preserving those components with a cumulative variance greater than 95%. These components were used to train five regression models: Random Forest, Support Vector Regression (SVR), Multi-Layer Perceptron (MLP), Partial Least Squares Regression (PLSR) and XGBoost. The performance of the models was evaluated using the RMSE, MSE and coefficient of determination (R2) metrics, applying cross-validation to improve the robustness of the results. The Random Forest model achieved the best performance for the 'Hass' variety, with an R2 of 0.8791, while for the 'Strong' variety, the XGBoost model obtained the best result with an R2 of 0.7113.

Instituciones: Universidad Tecnológica del Perú