Predicting Land Suitability for Wheat and Barley Crops using Machine Learning Techniques

Abstract

Agricultural land suitability evaluation is among the agricultural activities intended to enhance productivity. However, ensuring food security required to meet the demand for growing population is remain a key challenge across the globe particularly for developing countries like Ethiopia. This is due to most of existing study attempted to identify the level of land suitability for various crop is based on conventional which consumes time, cost, effort, and unable to predict the level of land suitability quickly and accurately. Undoubtedly, land suitability assessment with machine learning techniques can overcome this problem and enhance crops productivity. Therefore, this study is aimed to predict land suitability using machine learning techniques to enhance the productivity of the two most commonly cultivated cereal crops in Ethiopia; Wheat and Barley, using dataset collected from the Engineering Corporation of Oromia and it undergoes several preprocessing steps to make it ready to build machine learning models. Machine learning models such as Random Forest (RF), Gradient Boosting (GB), and K-Nearest Neighbor (KNN) with feature selections methods like Univariate Feature Selection (UFS), Recursive Feature Elimination with Cross-Validation (RFECV), and Sequential Forward Selection (SFS) methods are experimented in this study, and results are compared to identify the most performing model. Min-max feature scaling is applied to our dataset to build KNN model. However, there is no performance improvement exhibited. Hyperparameters are tuned to further optimize the performance of the models hence cross-validated randomized search hyperparameter tuning algorithm was used. The performance of the models with and without feature selection is evaluated under stratified 10-fold cross-validation with performance metrics such as accuracy, precision, recall, and f1-score including confusion matrix are used to compare the model's performance. The performance evaluation results revealed that the GB model with SFS method best performs than other models with and without feature selection since it scored an accuracy of 99.41%, precision of 99.37%, recall of 99.34%, and f1-score of 99.35%. Therefore, GB with SFS model is recommended to effectively predict the level of land suitability for wheat and Barley crops productivity. Besides what this study contributes in developing and evaluating machine learning models to predict land suitability for both crops, it is better to extend to a deep learning approach by incorporating a large dataset.