Prediction of mango firmness by near infrared spectroscopy tandem with machine learning

Abstract

The firmness of the mango fruit is one of the internal physical properties that can show its quality. Unfortunately, non-destructive methods to measure this are not yet available. In the current study, we develop a calibration model using near infrared spectroscopy to predict the physical properties (firmness) of the mango cultivar Arumanis (Mangifera indica cv. Arumanis) via machine learning. Spectral data were acquired using the fourier transform near-infrared (FTNIR) benchtop with a wavelength range of 1000 to 2500 nm. Multivariate spectra analysis based on machine learning, including principal component regression (PCR), partial least squares regression (PLSR), and support vector machine regression (SVMR), was utilized and compared to estimate the firmness of fresh mangos. The results obtained show that the prediction of machine learning by PLSR is better than that of SVMR and PCR for the prediction of mango firmness. The coefficient correlation of calibration (rc) and validation (rcv), the root means square error of calibration (RMSE-C) and validation (RMSE-CV), and the ratio of prediction to deviation (RPD) were 0.941, 0.382 kgf, 0.920, 0.472 kgf, and 2.556, respectively. The general results satisfactorily indicate that near infrared spectroscopy technology integrated with an appropriate machine learning algorithm has optimistic results in determining the firmness of mango non-destructively.