Blockchain-Enabled Modified Bi-Directional Recurrent Neural Network for Intelligent Crop Recommendation

Abstract

Global food demand is increasing rapidly, while variations in soil characteristics pose persistent challenges to sustainable agriculture. Inappropriate crop selection exacerbates chemical input requirements, accelerates soil degradation, increases cultivation costs, and reduces yield stability. To address these challenges, this paper proposes a data-driven decision-support system that integrates Internet of Things (IoT)-based soil sensing, edge-side data refinement, deep learning for crop recommendation, and blockchain to ensure security. The architecture includes: (i) a Sensor layer with custom-designed nodes to capture soil nutrient; (ii) an Edge layer implementing a Refine-and-Filter (RF) algorithm to eliminate irrelevant data, thereby improving accuracy and reducing communication overhead; and (iii) a Cloud layer employing a private blockchain for secure storage of edge layer data and a modified bidirectional recurrent neural network (modified BRNN), optimized using Gaussian Process Enhanced Hyperband with Neural Tangent Kernel (GPEH-NTK), to generate crop recommendations. Beyond the cloud layer, recommendations are stored on the InterPlanetary File System (IPFS), and their content identifiers (CIDs) are recorded on a separate public (Ethereum) blockchain. The public blockchain also provides decentralized user authentication. Experimental evaluation on public datasets achieved 98.85% accuracy, 98.92% precision, 98.88% recall, and 98.90% F1-score, surpassing state-of-the-art methods. Blockchain analysis confirmed tolerable latency without compromising security and integrity. A field deployment in Darbhanga, Bihar, validated real-time applicability. The findings suggest that the proposed system delivers secure and accurate crop recommendations and offers a viable solution to reduce cultivation costs, enhance yields, and promote sustainable food production.