RESEARCH · 2022-2023 · Applied machine learning
Real-time Indoor Positioning System.
Data Scientist & AI/ML Engineer
Cairo, Egypt
Source: resume + original portfolio + LinkedIn
Cairo, Egypt
Source: resume + original portfolio + LinkedIn
A BLE-based indoor positioning system integrated with the human activity recognition workflow to support patient-aware monitoring and event detection.
Algorithms tested
12
Supervised baselines
Best F1-score
84.12%
Random Forest
Integration
HAR + IPS
Unified monitoring logic
// 02 — CHALLENGE
WHY IT MATTERED
Indoor positioning is noisy by nature, so the project had to produce stable predictions from imperfect BLE signal behavior.
It also had to become useful in context, which meant integrating location with detected human activity rather than presenting location alone.
// 03 — APPROACH
HOW I BUILT IT
I tested multiple supervised models on BLE RSSI fingerprints and used ensemble thinking to make predictions more robust against fluctuating signal strength.
The system was connected with the activity-recognition pipeline to detect higher-level situations like prolonged immobility in risky locations.
// 04 — HIGHLIGHTS
KEY TAKEAWAYS
- →Used ESP32 receivers and RSSI fingerprinting for indoor location inference
- →Benchmarked 12 supervised learning algorithms
- →Deployed Random Forest as the strongest model with 84.12% F1-score
- →Integrated location and activity streams to trigger caregiver alerts
// 05 — OUTCOMES
RESULTS AND LESSONS
- →Built a practical machine learning view of indoor positioning rather than a purely hardware-centric prototype.
- →Connected model outputs to alerting logic that mattered for the care-facility use case.
- →Extended the overall monitoring stack from activity recognition into spatial awareness.
// 06 — STACK
THE TOOLS
Models
Random ForestHard VotingSVMGBMLDA
Hardware
ESP32BLERSSI fingerprinting
Apps
PythonFlaskFirebaseEmail alerting
// 07 — LINKS
SOURCE TRAIL