Hybrid filter and wrapper methods based feature selection for crop recommendation

The agriculture sector is certainly benefited from advances in science and technical proficiency, which has resulted in the generation of vast volume of data. With huge processing techniques, machine learning has enabled perception of new possibilities in agriculture management. Implementing machine learning algorithms over agricultural data is the significant onset for investigating and determining solutions to agrarian challenges. Agricultural data emerges from diverse sources such as weather, soil, crop characters and so on. Feature selection methods plays significant role to eliminate irrelevant features and identify significant features thereby enhancing machine learning model's performance. The paper proposes a hybrid feature selection strategy by combining filter and wrapper methods. The proposed method obtains optimal features from soil properties, crop characteristics, and climatic parameters in order build a crop recommendation model with better accuracy and performance. The effectiveness of the model is measured by considering all the features in the dataset and with the features obtained from proposed method. The proposed feature selection method is validated through evaluation metrics MSE, RMSE, MAE, R
²
. The machine learning models viz. artificial neural networks and decision tree are implemented with the selected features.