Improving Agricultural Efficiency: Severity-based Diagnosis of Bottle Gourd Leaf Diseases Using Federated Learning CNN

The identification and categorization of Bottle Gourd leaf illnesses across various severity levels are presented in this research as a unique application of the foundation of the study is a separate federation of five customers, each representing a different severity level, ranging from the lowest 1-20% severity range to the highest 81-100% severity range. The model consistently displayed good values across all customers and severity levels in investigating the local model performance (Results investigation). Client C1_j1 (1-20% severity), for example, reported an F1-Score of 85.57%, whereas client Cl_g5 (81-100% severity) attained an F1-Score of 95.17%, suggesting a high degree of accuracy and recall throughout all illness severity stages. The global aggregated for each client was used to analyze the model’s overall performance (Global Results Analysis). The lowest severity level (C1_j1) had an accuracy and recall of 85.27% and 85.94%, whereas the maximum severity level (Cl_g5) had a precision and recall of 92.88% and 92.69%. The metrics’ consistent rise as disease severity increased showed the resilience and accuracy of the model’s illness detection and classification skills over various disease severity levels. The macro, weighted, and micro average accuracy rates were 85.60%, 87.31%, and 87.26% for the least severe instances (C1_j1) and 92.78%, 93.16%, and 93.18% for the most severe cases (Cl_g5), respectively. These numbers demonstrate the model’s accuracy and toughness in the face of varied illness severity. It provides a significant resource for the agricultural sector, particularly for farmers in India and worldwide. This may result in early disease diagnosis and treatment, boosting crop production and total agricultural output.