Analysis of Phosphorus Content in Chickpea Samples using Near-infrared Spectroscopy and Machine Learning Models

This study aimed to develop and evaluate a rapid, non-destructive method for predicting phosphorus content in chickpea flour using near-infrared (NIR) spectroscopy combined with machine learning techniques. The research sought to identify the most effective wavelength range and modeling approach for accurate phosphorus content estimation.

NIR spectra was collected from 237 chickpea flour samples, of which 132 showed detectable phosphorus reflectance. The dataset underwent preprocessing techniques including noise elimination, data reduction and scatter correction. The interval Partial Least Squares (iPLS) algorithm was employed to identify the optimal wavelength range for phosphorus prediction. Five machine learning models-linear regression (LR), support vector regression (SVR), random forest (RF), decision tree regression (DTR) and neural network (NN)-were developed and compared using a dataset split into calibration (80%) and validation (20%) sets. Principal component analysis (PCA) was used for dimensionality reduction.

The most effective wavelength range for phosphorus prediction was identified as 1520-1658 nm. The Linear Regression model demonstrated the best performance on the testing data, with an R² of 0.945, Root Mean Square Error (RMSE) of 0.044 and residual standard error (RSE) of 0.045. All models except RF and DTR showed excellent predictive capability (RPD>4.1). The Decision Tree Regression model performed best on the validation set, with an R² of 0.875 and RMSE of 0.079. The study confirmed the potential of NIR spectroscopy combined with machine learning as a rapid and accurate method for phosphorus content prediction in chickpea flour.