In the Department of Biomedical Engineering, the defense of master’s student Rawan Latif Shkhair was held on Wednesday, August 20, 2025, regarding her thesis entitled:

Machine Learning for Automatic Preictal (Pre-Seizure) States of Epileptic Seizures in EEG Signals

The discussion committee consisted of:

• Committee Chairman: Prof. Dr. Ali Abd Al-Ilah Nouri

• Member: Asst. Prof. Dr. Hadeel Qasim Wadi

• Member: Lecturer Dr. Samar Ali Jaber

The thesis was supervised by: Asst. Prof. Dr. Ahmed Faiq Hussein – College of Engineering / Al-Nahrain University

The thesis was scientifically evaluated by:

• First scientific evaluator: Asst. Prof. Dr. Mohammed Sabah Jirjis – Northern Technical University – Technical Engineering College / Mosul

• Second scientific evaluator: Asst. Prof. Dr. Mohammed Sabri Salem – College of Engineering / Al-Nahrain University

The thesis was also linguistically evaluated by: Lecturer Dr. Basma Abd Al-Sahib Fayhan.

This study aims to develop an intelligent framework based on machine learning (ML) and deep learning (DL) techniques for automating the detection of epileptic seizures and predicting the preictal state using electroencephalogram (EEG) signals.

The proposed system relies on a one-dimensional convolutional residual neural network (ResNet1D) to extract distinctive features from raw signals, then pass them to lightweight traditional classifiers such as SVM and XGBoost to achieve high accuracy with minimal processing time.

The methodology included the following stages:

• Preprocessing of signals and applying filtering and segmentation techniques.

• Extracting deep features through CNN models and combining them with features from the time and frequency domain.

• Selecting the optimal features using the Random Forest algorithm to reduce dimensions by more than 95% while maintaining accuracy.

• Using dimensionality reduction techniques (PCA and t-SNE) to verify class separability.

Results
The proposed system achieved:

• AUC close to 0.9999 in seizure detection without false alarms.

• AUC of 0.9960 and F1-score = 0.981 in preictal state prediction.

• Reduced inference time to less than 10 milliseconds per EEG window (5 seconds), making it suitable for real-time applications on portable devices or clinical alarm systems.

The thesis was accepted for fulfilling the requirements for obtaining a Master’s degree in Biomedical Engineering.