<![CDATA[This study discusses the performance comparison of two optimization algorithms, Adam and RMS Prop, in training an Artificial Neural Network (ANN) for handwriting recognition classification tasks using the MOIST dataset. The MOIST dataset contains 70,000 handwritten digit images (0–9) measuring 28×28 pixels. The research stages include dataset input, preprocessing (pixel normalization, training and testing data division, and label conversion to one-hot encoding), weight initialization, and ANN model training with variations of Adam and RMS Prop optimizers. Model performance was tested using accuracy, precision, recall, and F1-score. The test results were carried out using the Google Cola application; the testing scheme was carried out in a structured manner with epoch iterations of 50, 100, 150, 200 and 250 on the Adam and RMS Prop optimizers. The test shows that the Adam optimizer generally provides higher accuracy and faster convergence than RMS Prop on the MOIST dataset, thanks to a combination of momentum and a more stable learning rate adaptation. However, RMS Prop still excels in weight update stability in some cases and tends to be more consistent on datasets with fluctuating gradient distributions.]]>
Copyrights © 2026
