Comparative Performance Analysis of IoT-Based Crop Recommendation Systems Using Random Forest and GridSearchCV Hyperparameter Tuning

Abstract

Optimal crop recommendation system development is essential for improving agricultural productivity through the assurance of compatible crop selections with soil and environmental factors. This study uses multi-sensor data collection from the Internet of Things (IoT) and machine learning approaches to design a specific and adaptive crop recommendation system. A publicly available dataset from Kaggle, containing 2,200 samples from 22 crop types with seven key attributes (N, P, K, temperature, humidity, pH, and rainfall), was used. The performance of several machine learning algorithms, such as Random Forest (RF), Decision Tree, Support Vector Machine (SVM), k-Nearest Neighbors (k-NN), and Linear Regression, was compared both with and without hyperparameter tuning using GridSearchCV. The outcome reveals that RF with GridSearchCV optimization achieves an optimal accuracy value of 99.24% for a 70:30 train-test split, outperforming other models in different performance measures and train-test partition ratios. A comparison of inference times further shows that the developed model achieves a satisfactory balance between accuracy and processing times, making it suitable for real-time IoT applications. System implementation and performance evaluation were performed on a Raspberry Pi 4B with a Coral accelerator and soil sensors integrated, thereby demonstrating its potential for realtime implementation. These findings underscore the significance of hyperparameter optimization and data-driven modeling in enhancing precision and versatility in precision agriculture approaches. Future research will aim to investigate larger datasets, reduce noisy sensor outputs, and incorporate additional environmental parameters—such as sunlight intensity and soil salinity—into the system for further improved recommendation accuracy under changing conditions.