Stnbl-A Crop Yield Prediction Framework for Assam Using Stacking and Blending Methodology

Abstract

Agriculture forms the backbone of Assam’s economy, employing nearly 60% of the state's workforce. One of the enduring challenges faced by the Assam agricultural sector is accurately predicting crop yields. Generally, the local farmers decide the cropping pattern based on their past experiences, thus avoiding the adoption of alternative crops due to economic uncertainties. In the recent literature, machine learning (ML) techniques for crop yield predictions have been applied to a few states across India. However, limited studies have considered the unique climatic conditions specific to Assam. To address this issue, this work introduces a structured, region-specific dataset and a crop yield prediction framework that is built on top of the newly generated dataset. This crop yield prediction framework, named as STnBL (Stacking and Blending) is developed considering specific climatic conditions of Assam and thus enabling more accurate and locally relevant outcomes. The performance of STnBL is analyzed using the coefficient of determination (R²) and runtime efficiency. The results highlighted that integrating individual models within an averaging ensemble model improved prediction accuracy from 60 % to 96%. Furthermore, the comparative analysis demonstrated that heterogeneous ensemble models, using stacking and blending, exhibited better performance compared to homogeneous ensemble approaches for crop yield prediction.