🧠 CogniScan – Real-Time Attention State Classifier

A machine learning-based system for detecting focused vs distracted mental states using facial expression data and raw pixel features.

---

📌 Project Overview

CogniScan is a multi-model machine learning system that classifies facial expressions into focused or distracted states using raw image data. It evaluates the performance of:

• 🎯 SVM (Support Vector Machine)
• ⚡ XGBoost Classifier
• 🧠 Convolutional Neural Network (CNN)

The goal is to compare classical machine learning with deep learning to detect real-time attention patterns—laying the foundation for an intelligent productivity assistant.

---

🚀 Features

• Preprocessing pipeline for image normalization and resizing
• Training & evaluation of 3 separate models
• Clean model comparison & metrics output
• Ready-to-use notebook for inference demos
• A modular codebase for future AU/gaze integration (e.g., OpenFace)

---

🛠️ Tech Stack

• Languages: Python
• Libraries:
  • scikit-learn
  • xgboost
  • tensorflow/keras
  • opencv-python
  • matplotlib, seaborn, numpy, pandas
• Tools: Jupyter, GitHub, VS Code

---

📁 Project Structure

CogniScan/
│
├── README.md
├── requirements.txt
├── .gitignore
│
├── data/                # Raw and preprocessed images (not uploaded)
├── models/              # Saved SVM, XGBoost, and CNN models
├── notebooks/           # Jupyter notebooks for training & demo
├── scripts/             # Python scripts for training & preprocessing
├── utils/               # Helper functions
└── assets/              # Demo GIFs, model performance plots, etc.

---

📊 Results

Model	Accuracy	F1-Score	Notes
SVM	-	-	Fast to train, small size
XGBoost	-	-	Strong generalization
CNN	-	-	Best performance, deeper model

Note: Results may vary slightly based on dataset splits and hyperparameters.

---

🧪 Getting Started

1. Clone the repo:

git clone https://github.com/your-username/CogniScan.git
cd CogniScan

2. Create virtual environment (optional but recommended):

python -m venv venv
source venv/bin/activate  # or venv\Scripts\activate on Windows

3. Install dependencies:

pip install -r requirements.txt

4. Run a notebook or training script:

jupyter notebook notebooks/compare_models.ipynb

---

🖼️ Demo

Example predictions made by each model using unseen facial images.

---

💡 Future Additions

• Integrate OpenFace AU features
• Add real-time webcam inference
• Build a Streamlit dashboard
• Add distraction heatmap overlays

---

👤 Author

Warren Dmello
💼 LinkedIn
📧 warrenlukedmello@gmail.com
🧠 Focused on AI for productivity and behavioral analysis.

---