Towards a Unified Approach for Crop Yield Prediction

Abstract

Agriculture is one of the fundamental occupations for majority of the countries in the world. Especially, in developing nations like India, the country is primarily driven by agriculture sector, where agriculture and its associated businesses are the backbone of the Economy making it the integral revenue generator. With technological advancements in the recent years, crop yield prediction has gained wide importance, and has shown to have significant impact on the revenue generated from agriculture in every season. Multiple factors influence crop yield prediction, which in turn makes it a non-trivial and challenging task. Despite many proposed works in the area, crop yield prediction lacks a unified solution. This paper brings out the need for a unified framework through a comparative study of standard algorithms and attributes. The algorithms considered are Linear Regression, Random Forest, K Nearest Neighbors (KNN) and Stochastic Gradient Descent (SGD). Our results show that Random Forest outperforms the other standard algorithms by showing 91.62% accuracy in crop yield prediction. Further evaluation is done where the attributes that affect the crop yield most are ranked according to their impact based on Mean Absolute Error (MAE). With this, we make a case for the need for a unified approach for crop yield prediction.