Authors: Assistant Professor Prajina V K, Assistant Professor Bhargavi M R
Abstract: Despite the technological advancements in agriculture, it continues to be vulnerable to climate change effects, and poor crop choice due to unfavorable conditions results in low yields and monetary hardship to farmers. This paper proposes a hybrid machine learning approach for crop recommendation that takes into account not only the weather forecast (rainfall, temperature, humidity) but also soil characteristics (pH, nitrogen, phosphorus, potassium). It consists of two stages: the Random Forest algorithm for feature selection and prediction followed by the XGBoost algorithm for correction of predicted values. Applying the approach to the data set of 50,000 crop images tagged by location for 15 main crops within a period of 10 years (2015-2025) in India, the hybrid algorithm reaches the accuracy level of 94.2% compared to Random Forest (89.3%), XGBoost (91.6%), SVM (84.2%), and KNN (81.5%). Rainfall and minimum temperature were recognized as crucial features by the algorithm. The proposed algorithm is implemented in a smartphone application for farmers that provides recommendations based on weather forecasts for the next 5 days, which allows increasing crop yields up to 20-30%.
