ADHD-Detection-using-EEG-Amplitudes

Predictive-Modeling-for-ADHD-Detection-using-EEG-Signal-Amplitudes

📌 Project Overview

This project presents an end-to-end deep learning pipeline for detecting Attention Deficit Hyperactivity Disorder (ADHD) using EEG (Electroencephalogram) signals. It integrates signal preprocessing, time-series modeling, and advanced neural network architectures to classify ADHD vs Non-ADHD subjects.

The approach explores multiple deep learning models to capture both spatial and temporal characteristics of EEG signals for accurate prediction.

---

🎯 Objectives

• Analyze EEG signal patterns associated with ADHD
• Preprocess and structure raw EEG time-series data
• Develop and compare multiple deep learning models
• Evaluate model performance for predictive classification
• Build a reliable ADHD detection system

---

📂 Dataset Description

• Multi-channel EEG signal recordings
• Large-scale time-series dataset
• Labeled data (ADHD vs Non-ADHD)
• High-frequency neurological signals

---

⚙️ Methodology

1. Data Preprocessing

• Noise filtering and signal cleaning
• Normalization and scaling
• Time-series segmentation

2. Feature Representation

• Channel-wise EEG structuring
• Transformation into model-compatible input shapes
• Sequence preparation for deep learning models

3. Model Development

The following deep learning models were implemented and evaluated:

• EEGNet
  Lightweight architecture specifically designed for EEG signal classification

• Convolutional Neural Network (CNN)
  Extracts spatial features from EEG signals

• Bidirectional LSTM (BiLSTM)
  Captures temporal dependencies in both directions

• Temporal Convolutional Network (TCN)
  Efficient for long-range sequence modeling

• ResNet (1D CNN variant)
  Enables deep feature learning with residual connections

• Inception1D
  Advanced architecture using multi-scale convolution filters for capturing complex signal patterns

---

📊 Model Evaluation

Models were evaluated using:

• Accuracy
• Precision
• Recall
• F1-Score
• Confusion Matrix

---

📈 Results

• Successfully classified ADHD vs Non-ADHD subjects using EEG data
• Progressive improvement observed across model architectures
• Advanced models like Inception1D captured complex signal patterns more effectively

---

📊 Visualizations

• Raw EEG signal plots
• Preprocessed signal comparisons
• Channel-wise EEG visualization
• Model performance plots (confusion matrix, metrics)

---

🛠️ Tech Stack

• Programming Language: Python

• Libraries & Frameworks:

  • NumPy
  • Pandas
  • SciPy
  • Scikit-learn
  • TensorFlow / Keras or PyTorch
  • Matplotlib
  • Seaborn
  • MNE (EEG processing)

---

▶️ How to Run

1. Install Dependencies

```bash pip install numpy pandas scipy scikit-learn matplotlib seaborn mne tensorflow