Solar Flare Prediction using SVM

Overview

This project implements a Support Vector Machine (SVM) model for solar flare prediction using different feature sets and datasets. The model is trained and evaluated with cross-validation and incorporates experiments for feature selection, data normalization, and no-shuffle validation.

Project Structure

data/
  ├── data-2010-15/
  ├── data-2020-24/
.gitignore
model.py
README.md
requirements.txt

Files and Directories:

• data/: Directory containing the datasets used for training and testing.
  • data-2010-15/ and data-2020-24/: These folders contain the dataset files for two different time periods.
• .gitignore: A Git ignore file to exclude unnecessary files from version control.
• model.py: Python file implementing the SVM model, data processing, and experiments.
• README.md: This README file.
• requirements.txt : File to install all dependencies

Dependencies

This project requires the following Python libraries:

• numpy
• pandas
• scikit-learn
• matplotlib
• seaborn

You can install the required dependencies using the following command:

pip install -r requirements.txt

Alternatively, you can manually install each package using pip:

pip install numpy pandas scikit-learn matplotlib seaborn

Usage

1. Load and Preprocess Data

Data is loaded from the data/ directory. The dataset contains feature matrices and corresponding labels for solar flare prediction.

The feature matrices are created from various sources of solar flare data:

• pos_features_main_timechange.npy, neg_features_main_timechange.npy
• pos_features_historical.npy, neg_features_historical.npy
• pos_features_maxmin.npy, neg_features_maxmin.npy

2. Feature Selection and Model Training

The SolarFlareSVM class is used to train the SVM model. It includes functionality for:

• Feature creation and selection (feature_creation)
• Preprocessing with standard normalization (preprocess)
• Cross-validation and performance evaluation (cross_validation)
• Calculation of the TSS score for each fold of the cross-validation (tss)

3. Experiments

The project includes three main experiments:

Feature Experiment

The feature experiment tests different feature combinations and identifies the best feature set based on the TSS (True Skill Statistic) score.

best_combination = feature_experiment()

Data Experiment

The data experiment tests the model's performance across different datasets and plots the TSS scores for each dataset.

data_experiment(best_combination=best_combination)

No Shuffle Experiment

This experiment disables data shuffling in the k-fold cross-validation and tests the model's performance based on the fixed order of data in data_order.npy.

no_shuffle_experiment(best_combination=best_combination)

4. Output

For each experiment, the model will output:

• The confusion matrix for each fold of the cross-validation.
• The TSS scores and average accuracy for each combination or dataset.
• Plots for the TSS scores across different feature combinations or datasets.

Example:

python model.py

This will run the feature experiment, data experiment, and no-shuffle experiment sequentially and plot the results.

Author

• Dev Mody
• McMaster University
• October 2024