/Fashion-MNIST-Classification-using-CNN-Neural-Networks

Fashion-MNIST CNN Classification: Accurately classify clothing images using CNNs. Data preparation, model training, hyperparameter optimization, and comparison with other models. Goal: precise categorization of fashion items through deep learning.

Primary LanguageJupyter Notebook

Fashion-MNIST-Classification-using-CNN-Neural-Networks

Project Description:

This project focuses on building a Convolutional Neural Network (CNN) model to classify clothing images in the Fashion-MNIST dataset. The objective is to accurately classify various fashion items based on their image representations using deep learning techniques. The project involves data preparation, model training, hyperparameter optimization, and comparison with other CNN models using transfer learning.

Getting Started:

Follow the instructions below to get started with the project:

  1. Download the Fashion-MNIST dataset.
  2. Install the necessary dependencies (e.g., Keras, TensorFlow).
  3. Execute the provided Python script or Jupyter Notebook.

Data Preparation:

  1. Download and load the Fashion-MNIST dataset.
  2. Perform data description, cleaning, and checking for missing values or duplicates.
  3. Visualize the data using suitable visualization methods to gain insights.
  4. Display sample images from the dataset to visually understand the fashion items.
  5. Conduct correlation analysis to identify relationships between variables, if applicable.
  6. Perform any required preprocessing operations on the data.
  7. Encode the categorical labels for training and evaluation.

Training a CNN Neural Network:

  1. Implement a LeNet-5 network for Fashion-MNIST digit recognition.
  2. Modify hyperparameters to achieve the best performance possible.
  3. Evaluate the model using 5-fold cross-validation.
  4. Plot the accuracy improvement using the adopted techniques.
  5. Plot the convergence curve for the LeNet-5 model.
  6. Comment on the potential reasons for the accuracy improvement (if observed) or the lack thereof.
  7. Explore two other CNN models using transfer learning and compare their results with the fully trained LeNet-5 model.