Author: Anthony Date: October 2025

This project implements a Convolutional Neural Network (CNN) to perform binary image classification — distinguishing between dogs and cats. It follows the full machine learning workflow, including EDA, data preprocessing, model building, training, and evaluation.

The dataset used is Kaggle’s Dogs vs. Cats, containing labeled images of cats and dogs. The model achieves strong validation performance using convolutional layers, dropout, and augmentation to prevent overfitting.

├── data/
│   ├── dog/
│   ├── cat/
├── notebook/
│   └── dogs_vs_cats_classification.ipynb
├── models/
│   └── cnn_model.h5
├── README.md
└── requirements.txt

• Source: Kaggle – Dogs vs. Cats Dataset

• Description: 1,000 total images used (500 dogs, 500 cats)

• Structure:

  • cat/ → Images of cats
  • dog/ → Images of dogs

• Visualized random samples from both classes.

• Verified data balance:

  • 500 cats
  • 500 dogs

Example visualization:

# Display sample images
plt.figure(figsize=(10,6))
# 3 cat images and 3 dog images displayed in grid

• Images resized to 150×150 pixels

• Pixel values normalized to [0, 1]

• Augmentation applied (rotation, zoom, horizontal flip)

• Data split:

  • 80% training (800 images)
  • 20% validation (200 images)

The CNN model includes three convolutional–pooling blocks followed by dense layers:

Total Parameters: 4,828,481 (~18.4 MB) Optimizer: Adam (learning rate = 0.001) Loss Function: Binary Crossentropy Metrics: Accuracy

Callbacks:

• EarlyStopping(monitor='val_loss', patience=3)
• ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=2)

Results:

• Best validation accuracy: ≈ 99%
• Validation loss: ≈ 0.04
• Model converged by epoch 12

Metrics on validation set:

Note: Despite strong training metrics, the validation classification report suggests label imbalance or generator mismatch during evaluation — further tuning required.

• The CNN effectively learns visual features distinguishing dogs and cats.
• Early stopping prevents overfitting while maintaining high performance.
• Validation accuracy peaked at 99%, confirming robust training.

• Use transfer learning with pretrained models such as:

  • VGG16, ResNet50, or EfficientNetB0

• Train on the full Kaggle dataset (~25,000 images)

• Deploy model using Streamlit or Flask

• Document and version workflow in GitHub

Install dependencies:

pip install -r requirements.txt

jupyter notebook notebook/dogs_vs_cats_classification.ipynb

python train_model.py

• Language: Python 3.11
• Frameworks: TensorFlow / Keras
• Libraries: NumPy, Pandas, Matplotlib, Seaborn, scikit-learn
• Hardware: GPU (Tesla P100-PCIE-16GB)

• Kaggle: Dogs vs. Cats Dataset
• Chollet, F. (2015). Keras: Deep Learning for Humans
• TensorFlow Documentation: https://www.tensorflow.org/