SWaT Cyber-Physical Security Detection.
Cairo, Egypt
Source: resume + original portfolio + LinkedIn
A research project on the Secure Water Treatment dataset combining time-series forecasting, binary attack detection, and attack-point identification for industrial control systems.
Cyber-physical systems require temporal understanding, not just static classification. The model has to capture process behavior over time and identify when that behavior stops making sense.
The project also needed to bridge anomaly detection and interpretable attack localization.
I worked on two complementary tracks: forecasting-based anomaly detection for deviations from normal behavior, and supervised classification for direct attack recognition.
The evaluation compared multiple recurrent and hybrid architectures to understand which trade-offs best fit the SWaT environment.
- →Benchmarked BiGRU, BiLSTM, and LSTM forecasting models
- →Built normal-vs-attack classification and attack-point labeling pipelines
- →Combined spatial and temporal modeling with CNN-LSTM and MLP architectures
- →Focused on real-time detection logic for operational environments
- →Identified stronger-performing sequential models for the forecasting task.
- →Built a fuller security-monitoring view by pairing anomaly detection with attack-point classification.
- →Strengthened my foundation in multivariate time-series security modeling.