Fog-Enabled Machine Learning Approaches for Weather Prediction in IoT Systems: A Case Study.

Area of Science:

• Environmental Science
• Data Science
• Meteorology

Background:

• Accurate temperature forecasting is vital for public safety, environmental risk management, and energy conservation.
• Forecasting is hindered in areas with inadequate governmental measurement infrastructure.
• IoT sensor networks offer a solution for data collection in data-scarce regions.

Purpose of the Study:

• To improve temperature forecasting accuracy in data-limited regions.
• To identify optimal real-time processing methods for large-scale sensor data.
• To ensure the reliability of temperature predictions.

Main Methods:

• Collected temperature, pressure, and humidity data using IoT sensor networks.
• Pre-processed data with Discrete Wavelet Transform (DWT) for feature extraction and noise reduction.
• Employed and compared three deep learning models: Wavelet-processed Artificial Neural Networks (W-ANN), Wavelet-processed Long Short-Term Memory Networks (W-LSTM), and Wavelet-processed Bidirectional Long Short-Term Memory Networks (W-BiLSTM).

Main Results:

• The Wavelet-processed Bidirectional Long Short-Term Memory (W-BiLSTM) model achieved the highest performance with 97% test accuracy and 2% Mean Absolute Percentage Error (MAPE).
• W-BiLSTM significantly outperformed W-LSTM and W-ANN models in predictive accuracy.
• Forecasts validated against Turkish State Meteorological Service (TSMS) data showed 94% concordance, confirming model robustness.

Conclusions:

• The W-BiLSTM model enables reliable temperature forecasting, overcoming limitations of insufficient governmental measurement infrastructure.
• This approach supports data-driven decision-making for environmental risk management and energy conservation.
• IoT sensor networks combined with advanced deep learning offer a scalable solution for critical environmental monitoring.