👁️ Autism Detection from Eye Gaze Pattern Images

This project explores deep learning and classical machine learning approaches to classify individuals as Autism Spectrum Disorder (ASD) suspects or Typical Controls (TC) using eye gaze pattern images. A dataset of 547 samples was used to evaluate three distinct pipelines.

---

📌 Objective

To compare different machine learning pipelines for detecting ASD from eye gaze data, and determine which methods generalize best under constrained data conditions.

---

📊 Dataset

• Total images: 547
• Classes:
  • TSImages-1 (ASD suspects)
  • TCImages-0 (Typical controls)
• Split: 80% Train / 10% Validation / 10% Test
• Input: Eye gaze pattern images derived from eye-tracking data

---

🔍 Methods

✅ Pipeline 1: Fine-tuned CNNs

• Models used:
  • Modified CNN
  • ResNet50
  • Xception
  • EfficientNetB7
  • InceptionV3
• Best: ResNet50
  • Accuracy: 89.47%
  • F1 Score: ≈ 0.89

---

✅ Pipeline 2: Hybrid + Chaos Butterfly

• A hybrid architecture combining Xception and ResNet50
• Hyperparameters optimized using Chaos Butterfly metaheuristic
• Accuracy: 87.72%
• F1 Score: ≈ 0.88

---

✅ Pipeline 3: ResNet50 Features + SVM (Classical)

• Bottleneck features from ResNet50
• Linear Support Vector Machine classifier
• Best Overall:
  • Accuracy: 91.23%
  • F1 Score: ≈ 0.91

---

⚙️ Techniques Applied

• Real-time data augmentation: rotation, flipping, brightness, zoom
• Class-weighted loss to address imbalance
• Hyperparameter tuning with Chaos Butterfly metaheuristic

---

🧰 Tech Stack

• Python 3
• TensorFlow / Keras
• Scikit-learn
• Keras Tuner
• Chaos Butterfly Optimization (custom implementation)
• Matplotlib, Pandas, Seaborn

---

🚀 Getting Started

1. Clone the repo

git clone https://github.com/21amY26/Autism_Detection.git
cd autism-eye-gaze

2. Create and activate a virtual environment

# Create virtual environment
python -m venv venv

# Activate it
# On macOS/Linux:
source venv/bin/activate
# On Windows:
venv\Scripts\activate

3. Install essential packages (if running locally)

pip install tensorflow keras keras-tuner scikit-learn pandas matplotlib seaborn split-folders

4. Run the notebook

jupyter notebook autism_eye_set.ipynb

Or run directly in Colab