Fashion MNIST classification

Keywords: Fashion MNIST

Background

The Fashion MNIST dataset is a widely used dataset in the field of computer vision and pattern recognition. It consists of a collection of grayscale images depicting various fashion items such as shirts, dresses, shoes, and bags. In this study, the author explores the application of Artificial Neural Networks (ANN) for fashion image classification using the Fashion MNIST dataset. The grayscale images in the dataset are represented as 28x28 pixels, each containing intensity values ranging from 0 to 255. The author preprocesses the dataset by normalizing the pixel values to a range of 0 to 1, allowing for easier convergence during training. After that, the author splits the dataset into training and testing sets, with a ratio of 80:20, respectively and passes it into an ANN model with two hidden layers. The activation functions used in the hidden and input layer is ReLu, while SoftMax was used in the output layer, as there are multiple different classes of images. Batch Normalization was used in each layer, and the optimizer used was adam, the loss applied was of categorical cross entropy. After applying all these parameters, the accuracy of the test data came out to be 88 percent.

Aim & Objectives

Aim
To Classify the Fashion MNIST Using DL Techniques
Objective
Implement an ANN to achieve accurate classification of Fashion MNIST dataset with optimized training.
Explore ANN architectures for improved Fashion MNIST classification performance and robustness.
Evaluate the ANN model's generalization on Fashion MNIST, considering efficiency and scalability for real-world use.

Research Methodology

ANN Technique used

Software & Tools

Jupyter, Google Collab

— Preview truncated. The complete document (full chapters, references, diagrams, and appendices) is shared with clients as part of project delivery. —