“Hybrid SNN–ANFIS Framework for Predicting Crop Yields Under Climate Change Scenarios: a Case Study of Maharashtra, India"

Abstract

Precise estimation of the crop yield in the context of changing climatic conditions is the key issue to the food security and the agricultural decision-making. This paper presents a hybrid model which uses semiparametric Neural Networks (SNN), combined with Adaptive NeuroFuzzy Inference System (ANFIS) to predict crop yields of a set of chosen districts in Maharashtra, India. The model uses the past data (2000 2022) which includes data about climate variables (rainfall, temperature, solar radiation, and the level of CO2 in the atmosphere) together with local crop yields. The first was that the dataset was processed and standardized and a geospatial mapping of the study area was performed to contextualize local differences. Trained and tested hybrid SNN ANFIS against machine learning baselines, such as SVR, ANN, CNN-RNN, and compared based on RMSE, R2, and MAPE, the hybrid SNN ANFIS had better performance. Fuzzy inference is optimized with Gaussian membership functions, and the interpretation of the model was improved using feature importance analysis and Partial Dependence Plots (PDP). Generalizability of the model was measured by depicting spatial heatmap that shows that the prediction accuracy was similar across different districts thereby confirming prediction consistency. The findings indicate that rainfall and temperature have been found to be the most statistically significant factors when it comes to determining the yield with the hybrid model attaining %92 accuracy in the test data. In addition to enhancing the accuracy of the prediction, the suggested framework is suitable in real life agri-policy formulation and climate adaptation approaches because it promotes explainability.