1Department of Computational Intelligence, SRM Institute of Science and Technology, Chennai-603 203, Tamil Nadu, India
*Corresponding Author: M. Meenakshi, Department of Computational Intelligence, SRM Institute of Science and Technology, Chennai603 203, Tamil Nadu, India. Email: meenaksm6@srmist.edu.in
The future of agriculture is heavily dependent upon the wise use of technology, as modern farming transitions towards data-informed decision-making. Traditional agronomy relies heavily on heuristics and experience-based decisions-a methodology that is increasingly inefficient and inconsistent when scaling to meet global agricultural demands. This paper provides a smart farming framework that embraces several data-centric models for precision agriculture to enhance productivity and sustainability.
The study integrates three key machine learning models. First, crop yield is predicted by a random forest model using the agricultural crop yield in Indian States (1997-2020) dataset. Second, soil types are classified using K-means clustering based on Nitrogen, Phosphorus, Potassium and pH levels from the Soil Measures dataset. Third, a convolutional neural network (CNN) trained on a representative subset of 10,849 images selected from the PlantVillage dataset (approximately 50,000 labelled images) is employed to identify 38 distinct classes of crop diseases.
The random forest model achieved a high predictive accuracy with an R2 score of 0.9797 and a mean absolute error (MAE) of 9.47 kg/ha. The CNN model demonstrated robust performance with an overall accuracy of 87.69% and a macro average F1-Score of 0.8363 in disease detection. The K-means algorithm successfully segmented soil profiles, achieving a silhouette score of 0.3647. The integrated system minimizes input wastage and provides actionable insights for sustainable cultivation.
Crop yield prediction, Deep learning, Machine learning, Precision agriculture, Soil classification, Sustainability
