🧠 NOTE:
There are four feature extractors used in this project:

1. Time Domain

2. Frequency Domain

3. Continuous Wavelet Transform (CWT)

4. Discrete Wavelet Transform (DWT)
   
   

Each extractor generates multiple features from the EEG signals.

✅ Feature selection is performed using the Mutual Information (MI) score, which measures the relevance between each feature and the class labels.
📊 The MI scores are visualized in a bar graph to help identify the most informative features.

🎯 The selected features are then used as input for Machine Learning (ML) models.

🔁 Mutual Information (MI) — Recommended!

Measures any relationship (linear or nonlinear) between a feature and class.

Works for both classification and regression.

Recommended MI Score Vs Significant feature

0.1 🔥 Strong influence — keep it
0.05 – 0.1 ✅ Moderate — likely useful
0.01 – 0.05 ⚠️ Weak — test it with model accuracy
< 0.01 ❌ Very weak — likely irrelevant, consider dropping
= 0 🚫 No contribution — drop it

Time Domain Extractor

Frequency Domain Extractor

Continuous Wavelet Transform  Extractor

Penerangan:

1. This link launches the Streamlit dashboard for extracting and selecting CWT features from raw EEG signals.

2. The raw EEG signals are preprocessed as follows: artifacts are removed using the ICA algorithm, the signals are segmented (the number of segments is user-defined), frequency bands (Alpha, Beta, Theta, Gamma, Delta) are extracted, and the bands are analyzed using CWT (the type of mother wavelet is user-defined).

3. The statistical features extracted after scanning the frequency bands are then normalized.

4. The extracted statistical features are then mapped to the classes using the Mutual Information (MI) score.

5. The statistical features include energy, entropy, mean, standard deviation, and others.

6. The relevant features can be downloaded for the development of a machine learning (ML) model.

Discreate Wavelet Transform  Extractor

Penerangan:

1. This link launches the Streamlit dashboard for extracting and selecting DWT features from raw EEG signals.

2. The raw EEG signals are preprocessed as follows: artifacts are removed using the ICA algorithm, the signals are segmented (the number of segments is user-defined) and the brain signals are converted to frequency domain. The signal are decomposed to 4 levels (Level 0, 1,2,3,4). At each level mother wavelet scan the lower frequency signals, and the statistical features are extracted.  (the type of mother wavelet is user-defined).

3. The statistical features extracted after scanning the frequency bands are then normalized.

4. The extracted statistical features are then mapped to the classes using the Mutual Information (MI) score.

5. The statistical features include energy, entropy, mean, standard deviation, and others.

6. The relevant features can be downloaded for the development of a machine learning (ML) model.