Comparative Evaluation of CNN, SVM, ANN, and kNN for Automated Detection of Nutrient Deficiencies in Maize Leaves

Resumo

Accurate detection of nutrient deficiencies in maize (Zea mays L.) is critical for optimizing yield and sustainable fertilizer management. Traditional diagnostic methods, including laboratory analysis and visual inspection, are often time-consuming and impractical for real-time applications. This study presents a comparative evaluation of four machine learning algorithms: Convolutional Neural Network (CNN), Support Vector Machine (SVM), Artificial Neural Network (ANN), and k Nearest Neighbors (kNN), for automated classification of nitrogen, phosphorus, potassium deficiencies, and healthy leaves. A labeled maize leaf image dataset was used to train and assess model performance based on accuracy, precision, recall, and F1 score. Results show that CNN achieved the highest overall performance (accuracy = 0.794, F1 score = 0.794), demonstrating superior feature extraction and robust classification. SVM performed competitively (accuracy = 0.722), ANN achieved moderate results (accuracy = 0.679), while kNN showed the lowest effectiveness (accuracy = 0.483). Misclassification was most prominent between potassium deficient and healthy leaves, reflecting overlapping visual symptoms. The findings highlight CNN as the most reliable model for automated nutrient deficiency detection.