Smart Farming for Groundnut Yield Prediction Using IoT and Machine Learning

Abstract

The integration of IoT in agriculture has revolutionized crop production by enhancing productivity, quality, and efficiency while reducing labor costs and boosting farmer income. IoT sensors provide precise data on environmental, soil, and plant factors, critical for predicting crop yields. In this study, groundnut crops were cultivated in 20 pots and monitored using IoT devices over 120 days, generating 480 data instances. Parameters like temperature, soil moisture, and nutrients (nitrogen, phosphorus, potassium) were measured to track growth metrics. Machine learning models Multi-Layer Perceptron (MLP), K-Nearest Neighbors (KNN), and Random Forest (RF) were developed using bagging techniques to predict yield and model growth rates based on NPK levels. Model performance was evaluated using R-squared, MAE, RMSE, and RMSLE metrics. For yield prediction, KNN outperformed RF and MLP with the highest R-squared (0.87), lowest MAE (2.1033), and lowest RMSE (2.0119), while MLP performed worst. Conversely, in modeling growth rates influenced by NPK, MLP excelled with the highest R-squared (0.52), lowest MAE (1.3499), MSE (2.7220), RMSE (1.6498), and an exceptionally low RMSLE (0.0024). Overall, KNN was the top performer for yield prediction, followed by RF and MLP, whereas MLP was superior for growth rate predictions. This highlights the potential of IoT and machine learning in advancing agricultural intelligence.