Minimax-Optimized Deep Learning Approach for Millet Yield Prediction Using an Improved LeNet CNN

Abstract

Crop genotypes, conditions, and agricultural techniques influence the environmental crops that yield forecasts. The present investigation used large datasets from weather conditions, soil, and crop characteristics to examine how effectively machine learning and deep learning algorithms can forecast millet harvests. We propose the use of an "Ample LE_ NET CNN(ALEN-CNN)" to utilize the convolution process to capture the time dependence of the environmental variables. ALEN-CNN is selected for its ability to model nonlinear relationships and capture spatiotemporal dynamics effectively, which are essential for accurate millet yield prediction under varying environmental conditions. In this instance, off-splitting normalization was performed to preprocess the information. Periodic features that might be retrieved using greedy trap fuzzy clustering and that specifically consider moisture levels in the soil and weather-related incidents were then used to address the seasonality ALEN-CNN was used. The proposed CNN performed better than the other machine learning models evaluated for predicting millet output in Tamil Nadu. Nonlinear models, such as deep learning models, have a stronger relationship between crop yield and input information than linear frameworks, and Deep Neural Networks outperform conventional methods in terms of prediction accuracy. The proposed ALEN-CNN model outperformed existing methods, achieving high accuracy (99.9%), F1 score (98%), and recall (94.6%) in millet yield prediction.