🐶 Dogs vs. 🐱 Cats - Image Classification Project

This project explores binary image classification for cats and dogs using three different modeling approaches:

1. ✅ Random Forest (with and without PCA)
2. ✅ Custom Convolutional Neural Network (CNN)
3. ✅ Pre-trained VGG16 using Transfer Learning + Fine-Tuning

---

📊 Dataset

We use a subset of the Kaggle Dogs vs. Cats dataset, containing:

• 1,000 training images
• 100 testing images
• Balanced classes (50% dogs, 50% cats)

---

🛠 Tools & Libraries

• Python 3.x
• TensorFlow / Keras
• scikit-learn
• matplotlib / seaborn
• NumPy / pandas

---

📈 Results Overview

Model	Validation Accuracy	Validation Loss
Random Forest (raw)	63.00%	-
Random Forest + PCA	59.20%	-
Custom CNN	72.00%	0.5461
Frozen VGG16	89.00%	0.2394
Fine-tuned VGG16	94.50%	0.1698

---

🧠 Key Insights

• Classical models like Random Forest perform poorly on raw pixel data.
• A Custom CNN improves performance but requires careful regularization.
• Transfer Learning with VGG16 achieves the highest accuracy and generalization, even with limited training data.

---

📂 Project Structure

• notebooks/ – Jupyter notebook with full model pipeline
• models/ – Saved Keras model files
• results/ – Accuracy and loss plots
• data/sample_images/ – A few example dog/cat images

---

🚀 Future Work

• Train on the full 25k Kaggle dataset
• Explore model ensembling (CNN + VGG16)
• Try ResNet50 or EfficientNet
• Improve training with early stopping, learning rate scheduling