A Machine Learning-Based Framework for Detecting Crop Nutrients Deficiencies.

Abstract

Efficient nutrient management is critical to enhancing agricultural productivity while promoting sustainable practices. However, traditional methods for diagnosing nutrient deficiencies in crops such as soil testing and visual inspection are often costly, time-consuming, and limited in scalability. This study proposes a machine learning-based framework for the automated detection of crop nutrient deficiencies, focusing on the three essential macronutrients: Nitrogen (N), Phosphorus (P), and Potassium (K). Utilizing a dataset of 1,156 leaf images, the system extracts 26 colour, texture, and shape based features. Feature selection methods, including ANOVA, Mutual Information, Random Forest, XGBoost, and Recursive Feature Elimination (RFE), are employed to identify the most discriminative features. Seven machine learning models are evaluated K Nearest Neighbours (KNN), Support Vector Machine (SVM), Naïve Bayes, Logistic Regression, Multi-Layer Perceptron (MLP), Random Forest, and Decision Tree using accuracy and weighted F1-score. The results demonstrate that feature selection significantly improves model performance, with Random Forest achieving 87.62% accuracy and MLP yielding the highest F1 score of 81.84%. This research highlights the potential of machine learning for scalable, real-time nutrient deficiency detection, contributing to the advancement of precision agriculture and sustainable farming.

Plant phenotyping relevance

葉画像から色・質感・形状特徴を抽出し、作物の栄養欠乏状態を自動推定する機械学習フレームワークの開発が中心であり、植物状態の取得・分類手法に該当する。

Code and data availability

The paper states it used a public Kaggle dataset ('Crop Nutrient Deficiency Detection Using Machine Learning', 1,156 leaf images) as its phenotyping input, but no Kaggle URL or repository identifier is provided in the supplied text, and no author analysis code, trained models, or data supplements are mentioned. No URL/

No evidence-backed public reproduction asset is currently recorded.