Crop Yield Prediction: A Semi-Supervised Self-Training Approach with Confidence Regularization

Crop yield prediction is essential for agricultural planning, resource allocation, and food security. Labeled data in the agriculture industry is typically costly and time-consuming, which is a crucial part of conventional machine learning models. In order to address the shortcomings of earlier models, this study utilizes a semi-supervised learning strategy that combines Self-Training and confidence regularization to improve agricultural yield estimation using both labeled and unlabeled data. The key issue addressed is the lack of labeled agricultural data, which limits the accuracy and generality of the model. Small-labeled datasets might cause overfitting of traditional supervised algorithms, resulting in inferior performance in various agricultural circumstances. In order to overcome this, the suggested technique labels the most certain predictions from unlabeled data using self-training. Confidence regularization dynamically adjusts prediction thresholds to use credible data, lowering the danger of error propagation and overfitting. The model performs well, with an accuracy of 88.66%, a precision of 90.65%, a recall of 90.00%, and an F1 score of 88.68%. Semi-supervised models outperform typical supervised models in contexts with minimal labeled data, as seen by these metrics. The suggested method improves crop production estimation and benefits farmers and policy makers by utilizing labeled and unlabeled data for real-world agricultural applications.