Hybrid CNN & Random Forest Model for Effective Guava Leaf Disease Diagnosis

This paper presents a detailed evaluation of a guava leaves disease categorization model, to contribute to the precise and early identification of numerous guava plant diseases. Precision, recall, F1-score, support, or accuracy metrics are used to evaluate the model's performance across multiple illness classes. Anthracnose, Rust, Powdery Mildew, Cercospora Leaf Spot, Bacterial Wilt, Guava Witches' Broom, Guava Whitefly, and Guava Leaf Miner are among the diseases represented in the collection. Individual disease classes get notable precision levels, with Anthracnose reaching 88.05% or Rust achieving 83.14%. Similarly, memory rates demonstrate the model's capacity to recognize positive occurrences, with Anthracnose recall rates of 74.87% and Guava Whitefly recall rates of 82.05%. The F1 scores offer a balanced picture of the model's precision versus recall trade-off, particularly high values indicating accurate disease detection. The macro, weighted, and micro averages provide information about the model's performance as a whole. The macro average produces an average precision of 76.73%, recalls of 86.00%, as well as an F1-score of 75.80% when each class is treated equally. The weighted average, when class distribution is taken into consideration, yields precision, recall, and F1-scores of 85.40%, 86.00%, or 84.90%, respectively. The micro average, which reflects a balanced approach, has a 74.87% precision, recall, and F1 score. The findings show the model's ability to discriminate guava leaf illnesses, arguing for its potential utility in overall disease management. The study emphasizes the need for proper disease detection for sustainable crop production, with the model provided as a promising tool for assisting farmers and specialists in limiting potential losses resulting from guava leaf diseases.