ML-CSFR: A Unified Crop Selection and Fertilizer Recommendation Framework based on Machine Learning

Abstract

Sustainable and substantial crop production is essential globally, especially considering the increasing population. To achieve this, selecting appropriate crops and applying necessary fertilizers are pivotal for ensuring satisfactory crop growth and productivity. Farmers have relied heavily on intuition when choosing which crops to cultivate and suitable fertilizers to use in a given season. However, this traditional approach often needs to consider the significant impact of current environmental and soil conditions on crop growth and yield. Overlooking these factors can have far-reaching consequences, impacting not just individual farmers and their households but also the entire agricultural sector. The integration of machine learning offers a promising avenue for addressing these challenges and providing practical solutions. The core contribution of this research lies in proposing a unified framework termed Machine Learning-enabled Crop Selection and Fertilizer Recommendation (ML-CSFR). This framework’s primary objective is to predict appropriate crops accurately and subsequently suggest corresponding fertilizers based on specific agricultural conditions. The initial phase involves the identification of proper crops for individual farmlands, considering local input variables. This phase employs artificial neural networks (ANN) to filter crops effectively using the available choices. The next phase utilizes soil and environmental parameters to anticipate the optimal fertilizer for the selected crops. This phase leverages the XGBoost (XGB) model to predict the most suitable fertilizers accurately. This two-phase approach ensures a comprehensive and effective recommendation system for enhancing agricultural outcomes. Experimental results demonstrate the effectiveness of this framework, achieving an accuracy score of 99.10% using ANN and 97.66% for XGB. The framework's capability to deliver tailored recommendations for individual farms and its potential to integrate real-time sensor data positions it as an effective tool for improving agricultural decision-making.