Folio — EditionSakina Saifuddin© 2026
2021

A journey through code, data, and intelligent systems.

Folio — EditionSakina Saifuddin© 2026
SS
01Intro02Work03What I Build04Achievements05Experience06Expertise07Contact
Case studyClick a page, swipe, or use ← → to turn

EEG SEIZUREDETECTION WITHDEEP LEARNING

EEG SEIZUREDETECTION

  • Classifies brain activity as seizure or non-seizure.
  • Studied the EEG signal before training.
  • Tested fold by fold on unseen data.
94%mean accuracy
5,508samples evaluated
Visit the portfolio
website at:sakina-saifuddin / works
27600
EEG Seizure DetectionSakina Saifuddin

EEG SEIZUREDETECTION WITHDEEP LEARNING

Tell seizures apart from normal activity
Fig. 1 — EEG channels behave differently.1
  1. Task
  2. My work
  3. Results
  4. Outcome

Tell seizures apart from normal activity

  • EEG data is noisy.
  • Channels can behave very differently.
  • The model must catch seizures, not just score well.
2
Signal first, model second
Fig. 2 — Statistics for every channel.3
  1. Task
  2. My work
  3. Results
  4. Outcome

Signal first, model second

  • Measured variance for every channel.
  • Checked mean, spread, skew and kurtosis.
  • Understood the signal before training.

Tools: Python

4
Tested on unseen folds
Fig. 3 — Loss falls, accuracy rises.5
  1. Task
  2. My work
  3. Results
  4. Outcome

Tested on unseen folds

  • Tracked training and validation every epoch.
  • Tested on held-out folds, not training data.
  • Compared results for each class separately.
0.9319fold 5 test accuracy

Tools: Deep learning · Cross-validation

6
Strong on both classes
Fig. 4 — Final classification report.7
  1. Task
  2. My work
  3. Results
  4. Outcome

Strong on both classes

94%mean accuracy
93%seizures caught
95%non-seizures cleared
5,508samples evaluated
8
Balanced, not just accurate
Fig. 5 — Stable training.9
  1. Task
  2. My work
  3. Results
  4. Outcome

Balanced, not just accurate

  • Similar results for seizure and non-seizure.
  • F1 of about 0.94 for both classes.
  • Academic work, not a clinical system.
10
11

Technologies

Python
Analysis and modelling.
Deep learning
Classifies the EEG signal.
Cross-validation
Tests on unseen data.
What I did
  • Analyzed the EEG channels and statistics.
  • Trained the classifier.
  • Ran the fold-level evaluation.
12

EEG SEIZUREDETECTION

  • Classifies brain activity as seizure or non-seizure.
  • Studied the EEG signal before training.
  • Tested fold by fold on unseen data.
94%mean accuracy
5,508samples evaluated
Visit the portfolio
website at:sakina-saifuddin / works
27600
Cover