Comparative Performance of ANN Predicting Soybean Yields from NDVI: Execution Time and CPU Usage in C and Python

Artificial Neural Networks (ANNs) are a widely used and powerful tool for modeling and predicting various processes across society, industry, and nature. In recent decades, ANNs have found increasing application in agricultural sciences, particularly in crop yield prediction. Despite their efficiency in solving such tasks, ANNs are known for their high computational demands. Most ANNs are developed using Python—a high-level, dynamically typed, interpreted language—which often results in less efficient use of computational resources compared to low-level programming languages. The primary objective of this study is to quantify the execution time gap between C-based and Python-based implementations of a simple feedforward ANN used for soybean yield prediction, utilizing Normalized Difference Vegetation Index (NDVI) values as input. Additionally, the study aims to evaluate and compare the predictive performance of both implementations. The results indicate no statistically significant difference in the accuracy of crop yield predictions between the C and Python models. However, a substantial difference in execution time was observed: the C-based ANN was found to be 21.1 times faster than the Python-based version. In addition, the C-based ANN saved CPU power, diminishing CPU usage 3.8 times. Therefore, we recommend using C for deep learning tasks in scenarios where optimal performance under limited computational resources is critical. Conversely, Python remains a suitable and preferable choice when ease of development is prioritized, and no substantial computational load is expected.