Explore the technical details of our hybrid CNN-LSTM architecture and IoT sensor integration that powers intelligent patient health monitoring.
Our model combines the spatial feature extraction capabilities of Convolutional Neural Networks (CNN) with the temporal sequence learning power of Long Short-Term Memory (LSTM) networks, enhanced by an attention mechanism that focuses on critical health events.
Multi-sensor time-series data (Heart Rate, SpO2, Temperature, Humidity, Air Quality)
Shape: (batch_size, time_steps, features) = (32, 60, 5)
Extract spatial correlations between different vital signs
3 Conv1D layers: 64, 128, 256 filters | Kernel size: 3 | Activation: ReLU | Max Pooling
Capture temporal dependencies and health trend patterns over time
2 Bidirectional LSTM layers: 128, 64 units | Dropout: 0.3 | Return sequences
Focus on critical time windows when health deterioration occurs
Self-attention layer with learned weights | Softmax activation
Final classification and prediction layers
Dense(64) → ReLU → Dropout(0.3) → Dense(3) → Softmax
Health status classification: Normal, Warning, Critical
3 classes with probability scores | Prediction confidence threshold: 0.85
OPTIMIZER
Adam (lr=0.001, β1=0.9, β2=0.999)
LOSS FUNCTION
Categorical Cross-Entropy
BATCH SIZE
32 samples
EPOCHS
100 (Early stopping: patience=10)
NORMALIZATION
Min-Max Scaling (0-1 range)
TIME WINDOWS
60-second sliding windows
SAMPLING RATE
1 Hz (1 reading per second)
FEATURE ENGINEERING
Rolling mean, std, gradients
ACCURACY
95%
INFERENCE TIME
0.12 seconds
MODEL SIZE
8.4 MB (TensorFlow)
PREDICTION HORIZON
30-60 minutes ahead
Main processing unit with WiFi/Bluetooth connectivity
SPECIFICATIONS
Heart rate and blood oxygen (SpO2) sensor
SPECIFICATIONS
Environmental and body temperature monitoring
SPECIFICATIONS
Detects harmful gases and air pollutants
SPECIFICATIONS
Complete sensor array with ESP32 microcontroller
$50
(₹4,000 approx.)
ESP32 connects to local WiFi network and transmits sensor data to cloud server via HTTPS/MQTT protocol.
Protocol: MQTT over TLS 1.2
Frequency: 1 Hz (every second)
Payload: ~200 bytes JSON
Local data logging to mobile app when WiFi unavailable, with automatic sync when connection restored.
Range: ~10 meters
Local storage: 24 hours buffer
Raw analog/digital signals from sensors
Basic filtering and validation on ESP32
Encrypted transmission to MongoDB database
Calculate statistical features and trends
CNN-LSTM model inference on processed data
Multi-tier notification based on predictions
POWER SOURCE
5V USB / Battery (18650 Li-ion)
CONSUMPTION
Average: 0.8W | Peak: 1.2W
BATTERY LIFE
8-12 hours continuous operation
SOLAR OPTION
5W panel for remote areas
WATCHDOG TIMER
Auto-reset on system hang
ERROR DETECTION
Sensor disconnect alerts
DATA BACKUP
Local 24-hour buffer storage
UPTIME
99.5% availability target
ENCRYPTION
TLS 1.2 for all communications
AUTHENTICATION
Device certificates & API keys
DATA PRIVACY
HIPAA-compliant storage
OTA UPDATES
Secure firmware updates