🤖 Human Action Recognition System

Advanced 15-class action classification using EfficientNetB0

This project implements a 15-class human action recognition system using transfer learning with EfficientNetB0. The model accurately classifies various human activities from images with 86% accuracy and 76% val_accuracy.

📦 Data Statistics

• Total images: 12,600
• Image dimensions: 224×224×3
• Training samples: 10,080 (80%)
• Test samples: 2,520 (20%)

🏷️ Activity Classes

• sitting
• using_laptop
• hugging
• sleeping
• drinking
• clapping
• dancing

• cycling
• calling
• laughing
• eating
• fighting
• listening_to_music
• running
• texting

🧠 Neural Network Structure

EfficientNetB0 (base model)
├─ GlobalAveragePooling2D()
├─ BatchNormalization()
├─ Dense(128, activation='relu', L2 regularization)
├─ Dropout(0.5)
├─ Dense(128, activation='relu')
├─ Dropout(0.3)
└─ Dense(15, activation='softmax')

⚙️ Training Configuration

• Optimizer: Adam (lr=1e-4)
• Loss: Categorical Crossentropy
• Batch Size: 32
• Base Model Frozen (except BatchNorm layers)

📥 Installation

1. Clone the repository:

   git clone https://github.com/your-username/15-Class-CNN-Classifier.git

2. Install dependencies:

   pip install -r requirements.txt

3. Download dataset from Kaggle and place in Data/ directory

🏋️ Training the Model

python train.py --epochs 60 --batch_size 32

######################################