Evaluating Machine Learning Models for Crop Recommendation with Optimal Nutrient Deviation Features

This research introduces a method for suggesting crops by combining domain-specific agricultural knowledge with machine learning models, focusing on the importance of deviations from optimal nitrogen (N), phosphorus (P), and potassium (K) levels to improve prediction accuracy. Feature engineering is used to compute deviation scores for N, P, and K, which are then applied in a comparative analysis of machine learning algorithms, like Random Forest, K-Nearest Neighbors, Logistic Regression and Support Vector Machine. The Random Forest algorithm has achieved a good test accuracy of 99.7% along with precision, recall, and F1 scores of 0.99. SVM, KNN, Logistic Regression and Decision Tree algorithms have achieved test accuracies of 97.95%, 97.50%,98.18%, and 88.64%, respectively. Additionally, a recommendation system was developed using nutrient ratios and cross-validation method in assessing nutrient levels in soil, helping farmers understand the percentage of soil nutrients. By focusing on both actual and optimal nutrient values, this research aims to help farmers in improving crop yield predictions.