hantrbl/cnn-lstm-temperature-forecasting

Hybrid CNN-LSTM neural network predicting temperature in Algiers from multivariate climate data. # CNN-LSTM Temperature Forecasting Undergraduate final project done for a Bachelor's degree in Operational Research. Accurate temperature forecasting has practical applications across energy planning, agriculture, and urban infrastructure. This project builds a hybrid CNN-LSTM model to forecast daily temperature in Algiers from multivariate climate data, achieving 1.94°C RMSE on an unseen test set. The model is lightweight (90K parameters), making it suitable for resource-constrained environments. **Key Features:** - Combines CNN for local pattern extraction with LSTM for temporal dependency modeling - Multivariate approach using 5 climate variables - 8 years of historical weather data (2016-2024) - Achieves **1.94°C RMSE** on unseen test set ## Results ### Performance Metrics | Metric | Value | |--------|-------| | **RMSE (Test)** | 1.94°C | | **MAE (Test)** | ~1.26°C | | **Relative Error** | 6.3% | | **Parameters** | 90,945 | ### Model Comparison Comparison of RMSE across different CNN-LSTM architectures ### Predictions vs Actual Temperature Time series comparison showing excellent tracking of seasonal patterns ### Prediction Accuracy Scatter plot demonstrating strong correlation between predictions and actual values ## Model Architecture ``` Input Shape: (timesteps, 5 features) ↓ Conv1D(256 filters, kernel=3, activation=relu) ↓ MaxPooling1D(pool_size=2) ↓ LSTM(64 units) ↓ Dense(32 units, activation=relu) ↓ Dense(1 unit, activation=linear) ↓ Output: Temperature (°C) ``` **Summary:** - **Total Parameters:** 90,945 - **Trainable Parameters:** 90,945 - **Optimizer:** Adam - **Loss Function:** Mean Squared Error (MSE) ## Dataset ### Overview - **Location:** Algiers, Algeria - **Period:** June 2016 - April 2024 (8 years) - **Frequency:** Daily observations - **Source:** rp5.ru Weather Archives) ### Features (Input Variables) | Variable | Description | Unit | |----------|-------------|------| | Temperature | Air temperature | °C | | Pressu …