Synthetic user profiling and deep learning for cold start mitigation in agricultural crop recommendation systems

Abstract

The cold start problem encountered in recommendation system where users have very little history so it becomes difficult to suggest crops is persistent in agricultural recommendation systems too. In this research paper, a new algorithm is suggested based on synthetic user profiling combined with deep learning to address the issue of data sparsity and enhance the quality of crop recommendations. The system is well designed such that it gives effective initial recommendations to new users or regions because of constructing synthetic profiles by clustering and mapping based on environmental similarity. The deep neural network (DNN) uses real as well as synthetic data to train the model and predict the optimum crop dependent on user and environmental inputs. Comparative simulations between the proposed DNN based approach and the traditional model like Random Forest, SVM, ANN, KNN and Decision Tree reveal that the offered model is higher in accuracy, recall, and F1-score. Remarkably, the model had generalization strengths on unseen profiles with high overall accuracy of 79 percent. It is the first systematic work on cold start problem in domain of crop recommendation formulated with the use of synthetic data augmentation and deep learning, which provides a scalable solution to precision agriculture in data-scarce settings.