An Empirical Evaluation of Optimal Deep Learning Models for Paddy Leaf Disease Analysis

Paddy is a significant food crop for most of the global population, but its yield is often reduced by leaf diseases such as Tungro, Hispa, Dead Heart, Brown Spot, and Blast. Detecting these diseases early is key to preventing crop damage and supporting sustainable farming. This study introduces an empirical evaluation of deep learning approaches to identify the comparatively best models for automatically classifying multi-ple paddy leaf diseases. In addition, an authentic dataset has been introduced, and five pre-trained models, such as VGG19, DenseNet121, InceptionV3, ResNet152V2, and MobileNetV2 have been appraised on the dataset. The performance of each model has been evaluated through familiar evaluation metrics, like accuracy, precision, recall, Fl-score, and ROC-AUC. The evalu-ation shows that the MobileNetV2 achieved as compared to best performance result with an overall 98.4 % of accuracy.