AI Predictions
Experimental LSTM · uses only reliable real sensor data
Updated: --
📍 Nizwa, Oman
Loading LSTM predictions…
Experimental LSTM
Next 6 Hours · Temperature Forecast
This is an experimental LSTM temperature prediction. The model uses real EcoCast sensor readings — input features: temperature, humidity, water level and light intensity. Pressure and altitude were excluded because the BMP390 sensor was faulty during deployment.
Training MAE: (reference only)
Horizon: 6 hours
Pressure and altitude excluded from the LSTM input — future work: replace the BMP390 sensor and add pressure back to improve prediction accuracy.
Peak
--°C
at --:--
Low
--°C
at --:--
Confidence
±--°C
over the 6-hour window
Hourly predictions · next 6 hours
One card per upcoming hour. Cards flagged red are outside the realistic 0–60 °C range.
Forecast curve · 6 hours
LSTM with confidence band that widens further out.
EcoCast LSTM
±error band
Academic honesty note. The LSTM was retrained on real EcoCast sensor data. Pressure and altitude were excluded because the BMP390 sensor returned invalid values during deployment. The Training MAE shown above is the model's score on the held-out training split — not a measure of live accuracy. The Live MAE appears only once the prediction_log has stored at least one Actual vs Predicted pair.
UTAS EcoCast · LSTM predictions update every hour via cron