Smart Disease Detection: Unifying CNNs and Random Forests for Accurate Pear Leaf Classification

The abstract summarizes their findings and presents a novel method for the automated detection and categorization of illnesses affecting pear leaves by combining Random Forest models and Convolutional Neural Networks (CNN). Using a carefully selected dataset of high-resolution photos, the model achieves impressive precision values, from 96.77% to 98.14%, demonstrating its capacity to reduce false positives for various disease classes, such as Leaf Spot, Powdery Mildew, Pear Rust, and Fire Blight. The model's ability to capture the majority of true positive occurrences is confirmed by the consistently high recall values, which range from 97.11% to 97.56%. This improves the diagnostic accuracy of the model. Moreover, the model's overall effectiveness is supported by the F1-Score, which is a harmonic mean of precision and recall with values ranging from 97.07% to 97.74% for various illness categories. This study evaluates the pear illness classification model using a dataset of 4746 photos. 97.34%, a remarkable total accuracy rate. The support values demonstrate this model's resilience over a large dataset, and the support proportion values underline the dataset's balance, which enhances the model's ability to generalize well. The model's overall success in disease classification is reflected in the weighted average and macro-average metrics, which offer a thorough assessment of the model's performance. The macroaverage is continuously high at 97.34%. This performance is mirrored by the weighted average, which corrects for class imbalance and validates the model's accuracy for a range of disease categories. The combined CNN and Random Forest method performed exceptionally well overall, taking into account both false positives and false negatives, as seen by the micro-average accuracy of 97.35%.