ABSTRACT

Indian farmers often rely on intuition and irrelevant factors, such as instant profits and misconceptions about soil potential, when making crop choices. This approach can lead to suboptimal agricultural practices and reduced productivity. To address this issue, a machine learning system can be employed to provide predictive insights about the best crops to grow. Highprecision machine learning algorithms can boost agricultural output and quality, promoting sustainable growth. This study focuses on recommending suitable crops using Land Resource Inventory (LRI) card data from the Hittinahalli sub-watershed in the Vijayapur district. Data was collected from the Sujala III Reward Project, managed by the University of Agricultural Sciences, Dharwad. The objective is to identify the most suitable crops based on the input provided by farmers. The model was trained using physical and chemical properties of soil, as well as weather parameters as independent variables, with crop suitability as the dependent variable. Two machine learning algorithms were utilized: Decision Tree and Random Forest. The performance of the models was evaluated using the F1-score, a measure of a test's accuracy. The Decision Tree algorithm achieved the highest accuracy with an F1-score of 0.88, compared to the Random Forest algorithm, which had an F1-score of 0.81. The prediction accuracy for specific crops varied: Cotton, Chilli, and Red Gram had an accuracy of 0.99; Sesamum, Groundnut, Jamun, Guava, and Cowpea had an accuracy of 0.92; and Banana, Mulberry, Lime, and Sapota had an accuracy of 0.78. The study demonstrates that machine learning can significantly enhance crop recommendation accuracy. This study contributes to the understanding of the complex connections between soil, weather, and crop success. By providing these insights, farmers and policymakers can make data-driven decisions to optimize yields and promote sustainable agriculture in the face of climate change.