Processing Near-Infrared Spectroscopy Signal to Calculate Soil Macronutrient: A Comparison of Some Machine Learning Approaches

Plantation plants need macronutrients Nitrogen (N), Phosphor (P), and Potassium (K) in the right amount during its growth period to grow optimally. Nowadays, the methods to measure soil macronutrient is still a relatively long time method, whereas it needs to be done as soon as possible. The rapid prediction system of the soil macronutrinet measurement is needed. The rapid prediction system requires the model of low energy consumption to apply on the portable mini computer. Most of the current researchs produce the model which requires the high energy consumption. This study aims to compare the performance of several machine learning (ML) algorithms in processing near-infrared spectroscopy (NIRS) signals with the application of several preprocessing techniques to get the best one to perform in the rapid predicition system to produce the low energy consumption model. Spectrum data is preprocessed using a random over-sampling technique and principal component analysis (PCA) of four number of principle components (PCs), 3, 5, 7, and 10. The training process of the model was conducted by four algorithms: random forest (RF), gradient boosting (GB), deep neural network (DNN), and partial least square (PLS). Prediction models on the N-P-K content data in mineral soils have been made, and it is found that the RF 10-PC algorithm is the most optimal for N with values of R
²
, RMSE, MAPE, NSE and Accuracy of 0.661, respectively; 0.022; 8,123; 0.667 and 91.88%. The RF 10-PC algorithm is the most optimal for P with evaluation parameters of 0.705; 9.250; 20.640; 0.806; and 79.36%. Also, the GB 3-PC algorithm is the most optimal for K with evaluation parameters of 0.774; 6.711; 10.307; 0.772, and 89.96%, respectively. This research found the low energy consumption model. PCA extract the features and also decrease the total of input variables from 131 to 10-PC and 3-PC of the model in order to decrease the model energy consumption.