Weather-Prediction-Model-using-LSTMs

This code implements a multi-time-step weather prediction model using LSTMs (Long Short-Term Memory Networks). It preprocesses weather data, creates time-series datasets, trains an LSTM-based neural network to predict future weather conditions.

Features

• Data Preprocessing:

  • Loads and processes weather data from a CSV file.
  • Splits data into training and testing sets for time series modeling.
  • Converts data into sequences for multi-step prediction.

• Model:

  • Implements a multi-layer LSTM neural network with dropout and fully connected layers.
  • Supports multi-output predictions over future time steps.

• Evaluation:

  • Uses RMSE, MAE, and R² metrics to evaluate the model on training and testing data.
  • Plots actual vs. predicted values for clear visualization of results.

• Visualization:

  • Displays prediction performance for each output variable using Matplotlib.

Dataset

The project uses a weather dataset (FullWeatherData_Toronto.csv) containing multiple input features (e.g., temperature, humidity) and numerical outputs for prediction.

Requirements

• Python 3.7+
• Required libraries:
  • numpy
  • pandas
  • torch
  • scikit-learn
  • matplotlib

Install dependencies using:

pip install numpy
pip install pandas
pip install torch
pip install scikit-learn
pip install matplotlib

Usage

Running the Code

1. Clone the repository:

   git clone https://github.com/<your-username>/time-series-weather-prediction.git
   cd time-series-weather-prediction

2. Place the FullWeatherData_Toronto.csv file in the project directory.
3. Run the script:

   python weather_prediction.py

Model Parameters

• lookback: Number of past time steps used as input.
• n_steps: Number of future time steps to predict.
• hidden_sz: Number of hidden units in the LSTM layers.
• dropout_prob: Dropout probability for regularization.
• n_epochs: Number of training epochs.
• batch_size: Batch size for training.

Outputs

1. Model Performance:
   • Logs metrics (RMSE, MAE, R²) for training and testing data at regular intervals.
2. Visualizations:
   • Plots actual and predicted values for each output variable in the dataset.

Example Results

• Example performance logs during training:

  Epoch 0: Train RMSE 5.4321, MAE 3.2145, R² 0.8721 | Test RMSE 6.1234, MAE 3.9876, R² 0.8567
  

File Structure

.
├── weather_prediction.py     # Main script for training and testing
├── FullWeatherData_Toronto.csv # Dataset file
├── README.md                 # Project documentation