Electronic sensing combined with machine learning models for predicting soil nutrient content

Abstract

Traditional methods for detecting soil nutrient content usually involve laborious and time-consuming experimental procedures, hindering the efficiency of soil analysis and making them less suitable for large-scale soil testing. Therefore, there exists a pressing need to develop innovative solutions. This study aimed to investigate the potential relationship between soil pyrolysis gas and soil nutrient content by establishing a model that combined electronic sensing and machine learning to predict soil nutrient content. The initial data sets were derived from utilizing pyrolysis gas response curves of soil samples acquired from 10 distinct gas sensors. Afterward, we innovatively proposed a sensor array optimization (SAO) method based on the dynamic feature importance (DFI) of the Random Forest-Pearson Correlation Coefficient (RF-PCC) to identify the optimal sensor combinations for four soil nutrients. Furthermore, Support Vector Machine (SVM), Random Forest (RF), Multilayer Perceptron (MLP), and Multilayer Perceptron-Random Forest (MLP-RF) models were employed to predict the soil nutrient content based on the electronic sense to address the challenges associated with developing a unified model. The MLP-RF models showed superior performance compared with the other three models, with the coefficient of determination (R2) for the four soil nutrients ranging from 0.54 to 0.94.Overall, our research not only provides approximately quantitative predictive outcomes but also contributes to a better understanding ofa new approach for detecting soil nutrient content.