A Comparative Analysis of AI-Based Smart Farming for Konkan Agriculture: Challenges And Future Direction

Abstract

Agriculture in the Konkan region faces challenges related to variable rainfall, soil conditions, crop productivity, and market uncertainties. This paper presents a comparative review of existing Artificial Intelligence (AI) and Machine Learning (ML) approaches for smart agriculture and evaluates a prototype Farmer Decision Support System designed for Konkan farming conditions. The proposed prototype integrates crop recommendation, Explainable AI (XAI), heuristic expected-yield estimation, and crop-variety recommendation. Four classification algorithms, namely Random Forest, Decision Tree, Support Vector Machine (SVM), and Logistic Regression, were evaluated using agricultural parameters including nitrogen, phosphorus, potassium, temperature, humidity, soil pH, and rainfall. Experimental results showed that Random Forest achieved the highest accuracy of 100%, with 100% precision, recall, and F1-score, followed by Decision Tree and SVM with 99.86% accuracy each, and Logistic Regression with 99.58% accuracy. SHAP-based explanations were incorporated to interpret crop recommendations. Historical yield analysis showed average yields of 1,579.82 kg/ha for Rice and 1,347.83 kg/ha for Ragi (Nachani), while the heuristic model estimated a Rice yield of 1,672.63 kg/ha under the evaluated conditions. The comparative analysis highlights limitations in existing approaches, including single-task functionality, infrastructure dependency, and limited regional adaptation. The findings demonstrate the feasibility of integrating AI-based agricultural decision-support modules while emphasizing the need for improved regional datasets, model validation, and field testing to support practical implementation in Konkan agriculture.

Keywords