Dogs vs. Cats Classification - Model Report

1. Model Architecture The model is based on MobileNetV2, a lightweight CNN designed for efficient image classification.

Model Layers & Design:

• Base Model: MobileNetV2 (pretrained on ImageNet, frozen during training)
• Custom Layers Added:
  • 3 Convolutional Layers (ReLU activation)
  • Global Average Pooling Layer (Reduces dimensionality)
  • Dense Layer (128 units, ReLU)
  • Dropout (50%) (Prevents overfitting)
  • Output Layer (Sigmoid activation) for binary classification (cat/dog)

Model Summary: Input Shape: (160, 160, 3)
Base Model: MobileNetV2 (Pretrained on ImageNet)
Custom Layers: Conv2D(32, 64, 128) → Dense(128) → Dropout(50%) → Sigmoid
Loss Function: Binary Crossentropy
Optimizer: Adam

2. Preprocessing Steps The dataset consists of 8,000 training images (4,000 cats + 4,000 dogs) and 2,000 test images (1,000 cats + 1,000 dogs).

Steps Used for Preprocessing:

• Image Rescaling: Normalized pixel values to range [0,1] (rescale=1/255)
• Data Augmentation (Training Only):
  • Rotation (20°), Zoom (20%), Width & Height Shift (20%)
  • Horizontal Flip (to improve model generalization)
• Resized Images to 160x160 Pixels (to reduce computation time)
• Binary Labels Assigned (0 = Cat, 1 = Dog)

---

3. Model Performance

• Training Accuracy: 96.86%
• Validation Accuracy: 94.22%
• Final Test Loss: 0.1411