Digital Cropping: A Comparative Analysis of Hybrid CNN-XGBoost With Stacking & Ensemble Techniques

Prediction of crop yield is crucial for farmers and governments worldwide to maximize profit and make informed decisions. Earlier investigations employed statistical methods but nowadays, Artificial Intelligence-based algorithms have emerged as practical tools for predicting the yield of crops. In this direction, the present study evaluates the effectiveness of various machine learning and deep Learning algorithms, including Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), Gradient Boosting (GBR), Bi-LSTM, LSTM, XGBoost, CNN-XGBoost, Stacking, Averaging, Bagging, and Linear Regression, in forecasting crop yield. The study was conducted on a dataset consisting of crop information about Rabi and Kharif season attributes with meteorological factors from 1997 to 2015 for different states in India. The models were compared and evaluated using various evaluation metrics. Our results showed that the Gradient Boosting method and stacking of machine learning techniques outperformed the existing baseline approaches with 98% prediction accuracy. Machine learning methods produced promising results in comparison to the regression model. By offering valuable insights into utilizing machine learning algorithms, our study contributes to crop yield prediction for diverse crops. The findings can assist farmers and stakeholders in making informed decisions about enhanced crop management practices.