Analysis of the Effects of Batch Size and Epoch Variations on MobileNetV2 for Mango Leaf Disease Classification

Mango (Mangifera indica L.) is one of the most widely cultivated horticultural crops due to its high economic value. However, leaf diseases remain one of the primary factors affecting crop productivity. Manual disease identification has several limitations, as it is time-consuming, highly dependent on the expertise of the observer, and inefficient for large-scale cultivation areas. This study aims to analyze the effects of epoch and batch size variations on the performance of the MobileNetV2 model for mango leaf disease classification using a transfer learning approach. The dataset consisted of 4,000 images representing eight classes: Healthy, Anthracnose, Bacterial Canker, Cutting Weevil, Die Back, Gall Midge, Powdery Mildew, and Sooty Mould. The dataset was divided into training, validation, and testing sets using an 80:10:10 ratio. Four training scenarios were then conducted using different combinations of batch sizes (16 and 32) and epochs (25 and 50 The experimental results demonstrated that the configuration with a batch size of 32 and 50 epochs achieved the best performance, obtaining an accuracy of 99.75%, while the precision, recall, and F1-score each reached 100%. These findings indicate that variations in batch size and number of epochs significantly influence the performance of the MobileNetV2 model, with the combination of a batch size of 32 and 50 epochs providing the most optimal classification performance for the mango leaf disease dataset.