Indirect Prediction of Dry Matter in Durian Pulp With Combined Features Using Miniature NIR Spectrophotometer

Immature durian poses a crucial problem for durian export. A miniature near-infrared spectrometer was used to develop a prediction model based on fruit stem and rind spectra. A sample of 120 durian fruits with variation in maturity was used to calibrate the model using different machine-learning algorithms, consisting of partial least squares regression (PLSR), least square support vector machine (LS-SVM), and artificial neural network (ANN) approaches. Initially, the rind model provided better predictive performance than the stem model for all algorithms. However, combining either the stem or the rind spectra (or even the fruit density) as the independent variables, the built models produced better results in prediction than from using either the stem or rind spectra alone. Among the models investigated, the combined spectra model of the stem and rind without spike spectra and using the ANN algorithm produced the best prediction with a correlation coefficient of 0.848, a root mean square error of prediction of 2.215%, and a residual predictive deviation of 1.833. Finally, the results showed that the proposed model performed sufficiently well in predicting the dry matter content of durian fruit using the miniature near-infrared spectrometer, even though it underperformed the model derived from the durian pulp.