Explainable Machine Learning for Crop Recommendation from Agriculture Sensor Data- a New Paradigm

The dwindling agricultural earnings and decrease in crop yield in recent years due to improper crop selection and fluctuation/ uncertainty in weather necessitate proper machine learning-based analysis. Machine learning methods can potentially alleviate the predicament caused by the lack of appropriate soil testing, consultation, and bias in manual suggestion. This work attempted to comprehend the agricultural sensor data and weather conditions and formulated the task in terms of supervised classification. The work obtained accurate suggestions in the presence of missing data, noise, etc. by using advanced machine learning methods. But recommendation alone is insufficient to convince farmers and other stakeholders to adopt this approach. Hence, this paper introduced explainable machine learning to completely comprehend the decision-making process. This work quantified the importance of features, explained individual prediction outcomes, and uncovered the rationale for decisions. The work employed state-of-the-art local interpretable model-agnostic, post-hoc explanation methods to provide in-depth insights. The insights obtained from the explanations can help the farmers develop a knowledge base and assist the farmers in choosing the appropriate sensors for the task. The human interpretable analysis enables the farmers to obtain satisfactory yields in these ever-changing and extreme weather conditions and environmental degradation.