AMR.ALFAYOUMY
← HOME / PROJECTS / REAL-TIME HUMAN ACTIVITY RECOGNITION
RESEARCH · 2022-2023 · Deep learning and IoT

Real-time Human Activity Recognition.

Data Scientist & AI/ML Engineer
Cairo, Egypt
Source: resume + original portfolio + LinkedIn

A wearable-sensor activity recognition system for elderly monitoring, including real-time inference, cloud-backed streaming, and a live dashboard.

Best F1-score
98.95%
CNN-LSTM
Best accuracy
98.95%
CNN-LSTM
Latency
<2 sec
Dashboard monitoring target
// 02 — CHALLENGE
WHY IT MATTERED

The work needed to balance modeling accuracy with enough latency and system design discipline for live use, not just offline evaluation.

It also needed to connect sensors, model inference, authentication, storage, and visualization into one usable flow.

// 03 — APPROACH
HOW I BUILT IT

I designed and benchmarked several deep learning architectures on sequential sensor data before settling on the best-performing hybrid CNN-LSTM setup.

The system was then wrapped in an IoT pipeline using live data streaming, cloud services, and a dashboard for near-real-time monitoring.

// 04 — HIGHLIGHTS
KEY TAKEAWAYS
  • →Benchmarked ANN, CNN, LSTM, and CNN-LSTM architectures
  • →Achieved 98.95% F1-score and 98.95% accuracy with CNN-LSTM
  • →Streamed live accelerometer data into a real-time monitoring workflow
  • →Built a dashboard for inference, metrics, and caregiver visibility
// 05 — OUTCOMES
RESULTS AND LESSONS
  • →Delivered a strong-performing activity recognition model with practical deployment thinking.
  • →Created a full-stack monitoring flow rather than a model-only experiment.
  • →Built a strong base for the later indoor-positioning integration work.
// 06 — STACK
THE TOOLS
Models
ANNCNNLSTMCNN-LSTM
Platform
FirebaseGCPJWT/OAuthStreamlit
Data
Sensor streamsPythonTensorFlow