Optimizing Crop Recommendations with Improved Deep Belief Networks: A Multimodal Approach

With the increase in world population, agricultural planning is significant to ensure food security. Timely recommendations for crops could be valuable for planning food production and maintaining food sustainability. This proposed work suggests a crop recommendation model considering physical soil characteristics, chemical soil characteristics, climate, and crop characteristics, using Improved Deep Belief Networks (IDBN). For this study, four important Indian crops—rice, maize, finger millet and sugarcane —were taken into account. The soil, weather, and crop datasets were gathered from different sources and subjected to feature selections method to extract optimal features. The crop recommendation model was developed using the proposed IDBN with Gaussian Restricted Boltzmann Machines and Ranger Optimizer with selected optimal features as input. Gaussian Restricted Boltzmann Machines enable the use of continuous values, and Ranger Optimizer ensures rapid convergence with selected optimal features as input. The findings from the investigation confirm that the suggested IDBN outperforms conventional Deep Belief Networks (DBN). The effectiveness of the crop recommendation model was evaluated using IDBN, and its performance was compared with several other machine learning algorithms.