Data-driven crop suitability assessment using soil health index and contaminant risk index in a multi-label learning framework
1Ahmedabad International School, Bodakdev, Ahmedabad, Gujarat 380054 India. 2Institute of Design, Nirma University, Ahmedabad, Gujarat 382481, India.Corresponding e-mail: ajay.goyal@nirmauni.ac.in
DOI: https://doi.org/10.37855/jah.2025.v27i04.121
Crop availability is determined by the physical, chemical, and biological properties of the soil, as well as environmental factors. The majority of crop recommendation models are primarily based on physical soil properties. Nonetheless, these approaches typically do not incorporate biological features, predict only one crop per sample, or use low-flexibility learning algorithms. To address the aforementioned limitations, this study introduces a biologically inspired machine-learning methodology for predicting the single and three most suitable crops based on soil-function indices (Soil Health Index and Contaminant Risk Index) calculated from microbial responses, soil structural properties, and contaminant markers. A dataset containing the physical and chemical properties of various types of soil was examined, which included soil samples collected from diverse agroecological regions across India, representing a broad range of soil types and climatic zones. The percentages of low SHI (62% degraded) and high CRI (66.66% contaminated) were 76.69 and 80.41, respectively. Supervised learning algorithms were implemented to handle a total of 13 single-crop responses. The model that combined Random Forest with Bagged Trees achieved 91.25% accuracy and an F1-score of 91.05%, with full (100%) test sample classification performance for the crops Coconut, Coffee, Rice, and Wheat. For multi-label classification, Bagged Trees achieved an exact match ratio of 84.00% and a Precision@3 of 79.08% in predicting the top three crops per sample. Errors occurred when crops had overlapping soil requirements (e.g., tomato and zucchini). This study may provide support for biologically guided crop selection in intercropping and a sustainable land-use system.
