A Hybrid Cost-Sensitive and Feature Reweighting Approach for Improved Crop Recommendation Systems

Accurate crop recommendation plays a crucial role in every agriculture-based country. Such recommendations are vital in regions where the choice of crops is affected in particular by several environmental factors. Recent trends and the techniques developed in machine learning often face challenges viz., class imbalance, where a few crops impact the dataset and the feature importance, that refers to certain environmental factors having an influence on the choice of crops than other several dependent factors. In this context, this paper proposes an innovative Hybrid Cost-Sensitive Support Vector Machine (SVM) model that uses Permutation-Based Feature Reweighting to handle both of these issues. The proposed model is tested and validated using the Kaggle Crop Recommendation Dataset, that contains 22 different crop types and 7 features of environment. To handle any class imbalance that arise due to unforeseen reasons this paper proposes a cost sensitive SVM, which use weighting technique to the most common crops, resulting in enhancement of the accuracy in prediction. Based on the enhanced experimentations a permutation-based feature re-weighting is suggested considering the environmental parameters such as pH and rainfall that effects the model accuracy and interpretability. Results show that the proposed technique achieve a precision of 97.73% and recall of 98.25% and F1-Score 97.81%, outperforming all the existing benchmarks.