<![CDATA[Maize represents a highly crucial agricultural staple within the Indonesian nation, where productivity is often affected by leaf diseases. Diseases like blight, common rust, and gray leaf spot prove hard to recognize by hand since the task demands time and risks personnel mistakes. Current research constructs a maize leaf disease classification model applying transfer learning based on InceptionV3 and evaluates the capabilities of three optimizing algorithms, specifically Adam, Stochastic Gradient Descent (SGD), and RMSProp. The dataset consists of 8,040 images collected from Kaggle and Mendeley, separated into four categories: blight, common rust, gray leaf spot, and healthy. Model training was executed using three dataset scenarios to assess the generalization ability of this suggested method. The empirical findings indicate that the Adam optimizer implemented on the merged dataset attained the highest effectiveness, reaching an exactness of 97.26%, as well as the highest precision, recall, and F1-score compared to SGD and RMSProp. The best-performing model was then converted to TensorFlow Lite and launched within a Flutter-driven Android software to help individuals with the initial spotting of maize leaf diseases.Keywords: Maize; Leaf Disease; Convolutional Neural Network (CNN); InceptionV3 AbstrakJagung adalah salah satu komoditas pangan penting di Indonesia yang kerap mengalami kendala produksi akibat serangan penyakit pada daunnya. Penyakit seperti hawar, karat, dan bercak daun abu-abu sukar dikenali secara manual karena memerlukan waktu yang lama dan berisiko mengalami kesalahan. Studi ini mengembangkan model klasifikasi penyakit daun jagung menggunakan transfer learning berbasis InceptionV3 dengan membandingkan kinerja tiga algoritma optimasi, yaitu Adam, SGD, dan RMSProp. Dataset yang digunakan terdiri dari 8.040 citra dari Kaggle dan Mendeley yang terbagi ke dalam empat kelas, yaitu blight, common rust, gray leaf spot, dan healthy. Pelatihan dilakukan pada tiga skenario dataset untuk mengevaluasi kemampuan generalisasi model. Temuan memperlihatkan bahwa algoritma Adam pada data kombinasi memberikan performa tertinggi dengan akurasi 97,26% serta nilai precision, recall, dan F1-score tertinggi dibandingkan SGD dan RMSProp. Model terbaik dikonversi ke TensorFlow Lite dan diimplementasikan ke dalam aplikasi Android berbasis Flutter sebagai alat bantu mengenali gejala awal penyakit tanaman jagung.Kata kunci: Jagung; Penyakit Daun; CNN; InceptionV3 ]]>
