Comparative Insights into Deep Learning-Powered Decision Support Systems for Crop Recommendation

Precision farming has become critical in modern agriculture, aiming to increase crop yield, sustainability, and resource efficiency through data-driven insights. Deep learning (DL)-based decision support systems offer promising advancements in this domain, enabling more accurate crop recommendations by integrating complex data, including soil conditions, climate, and crop histories. While various studies have examined various machine learning (ML) models as well ML in combination with DL for crop recommendation, there is a lack of comparative analysis focusing solely on different DL architectures. Addressing this gap, this study conducts a comprehensive comparison of four primary DL models - Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), Autoencoder (AE), and Deep Neural Network (DNN) - for crop recommendation. This analysis is structured around two critical dimensions: generalizability and performance, assessing each model's ability to adapt to new data and achieve high predictive accuracy. Results indicate that the LSTM model consistently outperforms other architectures in both generalizability and metrics like accuracy, precision, recall, and ROC AUC, making it highly suited for complex, time-sensitive agricultural data. This study provides valuable insights into the capabilities of different DL models, guiding future development of robust, DL-powered crop recommendation systems that support sustainable and precise agricultural practices.