HARIPS [Thesis].
Cairo, Egypt
Source: resume + original portfolio + LinkedIn
A combined human activity recognition and indoor positioning system for elderly monitoring, linking wearable-sensor inference with BLE location tracking and caregiver-facing alerts.
The core problem was not just recognizing movement or estimating location in isolation, but combining both into a monitoring system that could interpret context.
For an elderly-care setting, the valuable signal came from knowing what a person was doing, where they were, and whether that combination implied risk.
I built the HAR component around sequential accelerometer data, benchmarking ANN, CNN, LSTM, and CNN-LSTM models before selecting the strongest performer for real-time inference.
I paired that with an IPS pipeline based on BLE RSSI fingerprinting and supervised baselines, then integrated both outputs into one monitoring flow so location and activity could jointly drive caregiver-facing events.
- →Benchmarked deep learning models for human activity recognition and classical supervised models for indoor positioning
- →Reached 98.95% HAR F1-score with CNN-LSTM and 84.12% IPS F1-score with Random Forest
- →Used ESP32 receivers and BLE RSSI fingerprinting to add spatial awareness to the monitoring stack
- →Combined activity and location signals into higher-level caregiver alerts for risky situations
- →Turned two separate research tracks into one end-to-end monitoring concept rather than leaving them as isolated models.
- →Showed that context-aware monitoring improves when activity inference and spatial inference are interpreted together.
- →Produced a stronger systems story by linking model quality, IoT data flow, dashboarding, and alert logic in one design.