Soil Analysis-Guided Expert: A Machine Learning Device for Optimized Fertilizer Selection

Chemical fertilizers play a crucial role in enhancing yields and meeting the increasing demands in food production of growing populations. However, improper fertilizer practices can lead to diminished crop yields and adverse environmental repercussions. This research introduces the application of a machine learning device designed to aid in fertilizer selection. Utilizing Numpy, Pandas, and Scikit-learn libraries, Random Forest (RF), Decision Trees (DT), Support Vector Machine (SVM), XGBoost, and Neural Network (NN) algorithms were explored to identify the most fitting algorithm for integration into the device. The dataset used consists of 4513 samples, encompassing a single target variable (fertilizer) and nine features (district name, soil color, nitrogen, phosphorus, potassium, $\mathbf{p H}$, rainfall, temperature, and crop). Model optimization was performed using Optuna, and evaluation metrics included accuracy and $F 1$ score. The hardware implementation involved the use of Jetson Nano to meet machine learning requirements and Tkinter for Graphical User Interface development. Based on the results, the Neural Network emerged as the optimal algorithm, achieving a remarkable $\mathbf{9 9 \%}$ across all metrics. This research holds the potential to advance fertilizer selection practices, thereby fostering enhanced agricultural productivity while simultaneously promoting environmental sustainability.