Evaluation of Convolutional Neural Network Model Architecture Performance

The identification of image types can be facilitated through the use of algorithms and automation in their detection. In this research, detection is performed on fashion images, which encompass various types and categorizations. The process of identification and detection can utilize the Neural Network algorithm known as Convolutional Neural Network (CNN). The Convolutional Neural Network (CNN) algorithm consists of multiple layers, and its application involves setting values for pixels as well as arranging the sequence of layers within it. In this study, two Convolutional Neural Network (CNN) algorithm models are employed for image classification in the fashion dataset. These two models differ in the order of layers within the Convolutional Neural Network (CNN) and in the values of strides applied for modification. Model 1 employs ZeroPadding2D in the first layer with a smaller stride value of 1 compared to Model 2, which places the Conv2D layer in the first layer and applies strides with 1, 2, and 3 values. This research indicates that Model 1 outperforms Model 2, achieving the highest accuracy in epoch 20 with an accuracy value of 0.9711. This is attributed to the fact that the strides value in the Conv2D layer of Model 1 is set to a smaller value compared to Model 2.