- Introduction:
The primary aim was to develop a robust Convolutional Neural Network (CNN) for accurately classifying handwritten digits from the MNIST dataset. MNIST, a well-known collection of grayscale images representing digits 0 through 9, serves as an excellent testing ground for machine learning and neural network applications.
The project journey involved a comprehensive exploration of the dataset, where we delved into its distribution, visualized key samples, and meticulously preprocessed the data. The heart of the project lies in the design and optimization of a neural network using the Keras Sequential API. Through an iterative process, we fine-tuned the architecture and introduced regularization techniques to enhance the model's accuracy and robustness.
- Neural Network Design
- Model Compilation and Training
- Regularization Techniques (L2 and Dropout)
- Exploratory Data Analysis
- Visualization using Matplotlib
Workflow:
- Exploratory Data Analysis:
Imported necessary libraries and suppressed warnings for a cleaner output. Utilized Keras to import the MNIST dataset, a collection of handwritten digits. Examined the distribution of labels in both the training and test sets. Displayed a subset of 25 randomly sampled MNIST digits to gain initial insights.
- Data Preprocessing:
Calculated the number of unique labels in the dataset. Transformed class labels into one-hot encoded vectors. Determined the assumed square dimensions of the images. Flattened and normalized pixel values to a range between 0 and 1.
- Neural Network Architecture:
Constructed a Sequential model using Keras. Incorporated three dense layers with ReLU activation functions. Displayed a summary of the model architecture, including parameters.
- Model Training:
Specified the loss function, optimizer, and metrics for model compilation. Executed training for 20 epochs with a batch size of 128 and a validation split of 30%. Displayed training and validation accuracy and loss over epochs.
- Results and Evaluation:
Highlighted the achieved test accuracy of 97.2%. Plotted training and validation loss over epochs using Matplotlib. Evaluated the final model on the test set to ensure generalization. Displayed the test loss achieved by the model.
- Model Optimization - L2 Regularization:
Modified the model to include L2 regularization in the first dense layer. Observed changes in training and validation loss with the introduction of L2 regularization.
- Model Optimization - Dropout:
Enhanced model robustness by adding dropout layers after each dense layer. Analyzed changes in training and validation loss with the introduction of dropout.
- Combined L2 and Dropout:
Combined L2 regularization and dropout layers for a more resilient model. Executed training for 20 epochs with the combined regularization techniques. Evaluated the model on the test set and highlighted achieved accuracy.