Treffer: Enhanced deep neural network with interaction features for corn seed yield prediction: uncovering agroecophysiological relationships.

Context: Accurate prediction of seed yield is critical for optimizing agricultural practices, improving resource management, and understanding ecophysiological interactions in crop systems. This study focuses on maize seed yield prediction, leveraging advanced machine learning techniques to enhance prediction accuracy and support sustainable farming. Objective: The objective of this study is to develop an Enhanced Deep Neural Network (DNN) model that integrates interaction features to predict maize seed yield with improved accuracy and robustness compared to a baseline DNN model, while identifying key agroecophysiological factors influencing yield. Methods: An Enhanced DNN model was developed, incorporating interaction features to predict maize seed yield. The model underwent meticulous hyperparameter tuning and included early stopping with a patience of 50 epochs and a ReduceLROnPlateau callback (initial learning rate of 0.001, factor of 0.5, patience of 20) to prevent overfitting. Reproducibility was ensured by fixing random seeds for NumPy, Python's random module, and TensorFlow at 42 (i.e., np.random.seed(42), random.seed(42), tf.random.set_seed(42)). Sensitivity analysis was conducted using mean absolute weights of the first layer to evaluate feature importance. Residual analysis was performed using the Shapiro-Wilk test to assess the model's statistical reliability. Results and conclusions: The Enhanced DNN model achieved an R² of 0.483, RMSE of 2.794 t/ha, and MAE of 2.118 t/ha on the test set, significantly outperforming the baseline DNN model (R²=0.190, RMSE = 3.783 t/ha, MAE = 2.744 t/ha). Key features, such as RootColonization_SPAD1 and Stem Diameter, were identified as critical predictors, highlighting the role of nutrient dynamics and structural plant traits. The Shapiro-Wilk test (p-value = 0.070) confirmed residual normality, indicating no systematic bias and supporting the model's reliability. The incorporation of interaction features substantially improved variance explanation and prediction precision. Significance: This study demonstrates the transformative potential of interaction features and advanced DNN architectures in precision agriculture. By capturing complex agroecophysiological relationships, the model provides a foundation for developing accurate and generalizable predictive tools, supporting sustainable farming practices under diverse environmental conditions. [ABSTRACT FROM AUTHOR]

Copyright of Discover Agriculture is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)