EfficientNet-DigitClassifier-99Acc

Overview

This model card introduces the EfficientNet-DigitClassifier-99Acc, a high-accuracy Convolutional Neural Network (CNN) designed for digit classification. Achieving an impressive accuracy of 99.58%, this model stands out as a reliable tool for recognizing handwritten digits, trained and validated on a large-scale digit dataset with 240,000 samples for training and 40,000 for testing.

Model Architecture

The EfficientNet-DigitClassifier-99Acc features a sequential CNN architecture optimized for 28x28 pixel grayscale images. The architecture includes:

• Conv2D Layers: Two convolutional layers with 32 and 64 filters, respectively, using 3x3 kernels and ReLU activation.
• MaxPooling2D Layers: Pooling layers following each Conv2D layer to reduce spatial dimensions.
• Flatten Layer: Converts the 2D matrix data into a vector for processing in dense layers.
• Dropout Layer: A dropout rate of 0.5 to mitigate overfitting.
• Output Layer: A dense layer with 10 units and softmax activation, corresponding to the ten digit classes.

Dataset

Training and testing were performed on a specially curated digit dataset derived from the MNIST database, featuring 240,000 training samples and 40,000 test samples. Each image underwent preprocessing to scale pixel values to the [0, 1] range and reshape to 28x28 pixels.

Performance

• Test Loss: 0.0169
• Test Accuracy: 99.58%

This model exemplifies state-of-the-art performance in digit classification, providing a robust solution for applications requiring high accuracy in digit recognition.