A Performance Benchmarking of ML and DNN Models in Sorghum Yield Estimation for South India

Jayashree T R, Dr. NV Subba Reddy, Dr. U Dinesh Acharya

Abstract: The agriculture sector is extremely vulnerable to the consequences of climate change, leading to reduced quantity and nutritional quality of crop yields. The application of innovative techniques contributes to the modernization of agricultural practices and enhances productivity, quality, and food security. The accurate and reliable estimation of crop yields based on climate, soil, and crop-related data can guide/assist farmers in planning and managing agricultural activities in a better way. Nowadays, deep learning techniques are widely being explored for agricultural yield predictions due to their ability to address both linear and non-linear components of the data. The present work assesses the predictive capabilities of a tensor-flow-based deep neural network (DNN) in predicting yields of the jowar crop in the southern state of India. Comparatively better prediction accuracy was demonstrated by the DNN when compared to the performances of four other machine learning models. The DNN used the meteorological, soil, and crop data of the growing season of jowar for 15 years of duration and presented promising prediction results with R2 values of 0.88. The work incorporates the novel approach of using the SHAP (SHapley Additive exPlanations) framework to identify the contribution of the most significant feature towards prediction.

Keywords: Crop yield, Deep neural network, Jowar yield, Random Forest, Support Vector Regressor, Prediction

How to Cite?: Jayashree T R, Dr. NV Subba Reddy, Dr. U Dinesh Acharya, "A Performance Benchmarking of ML and DNN Models in Sorghum Yield Estimation for South India", Volume 14 Issue 8, August 2025, International Journal of Science and Research (IJSR), Pages: 161-170, https://www.ijsr.net/getabstract.php?paperid=SR25801151443, DOI: https://dx.doi.org/10.21275/SR25801151443