Sr.

No Paper Name About Paper

1 Prediction of Large Solar Flares This research paper investigates the prediction of large
Based on SHARP and HED Magnetic solar flares using machine learning. Two
Field Parameters datasets, SHARP and HED, containing different magnetic
field parameters from solar active regions are used to
train six models. The HED dataset, focusing on high-
energy density regions, shows superior performance
compared to SHARP, which uses parameters from the
entire active region. The Transformer model consistently
outperforms other models in both categorical and
probabilistic forecasting. Finally, a parameter importance
analysis identifies R VALUE (SHARP) and Efree (HED) as
the most impactful parameters for predicting large solar
flares.

This research paper investigates using machine learning

to predict powerful solar flares, which can severely
damage Earth's infrastructure. The study surveys various
machine learning models, using a newly available dataset
of solar flare data and remote sensing measurements
from multiple satellite missions (SOHO, SDO, ACE). The
MACHINE LEARNING MODEL goal is to develop a system of satellites with on-board
SURVEY WITH THE DATASET FOR machine learning capabilities to predict these events in
SOLAR FLARE PREDICTION advance, allowing for protective measures. The CatBoost
Classifier model performed best in the study, although
improving the recall (reducing false negatives) remains a
key area for future work. The research highlights the need
for advanced remote sensing and computationally
powerful satellites for early warning and mitigation of
solar flare threats.
 The researchers tested their ResNet model to predict
solar flares using HMI magnetograms (magnetic field
data) and compared it with other satellite data (AIA
images). They found that the model had 75.5% accuracy
with HMI data, much higher than the 71% to 71.3%
Extreme Solar Flare Prediction Using
accuracy with AIA data. This shows that HMI
Residual Networks with HMI
magnetograms, which focus on the Sun's magnetic field,
Magnetograms and Intensitygrams
are much better for predicting solar flares than AIA
images, which focus on the Sun’s atmosphere. The study
concluded that HMI magnetograms are more effective for
improving solar flare predictions and space weather
forecasting.

The paper presents a hybrid CNN model combining

Solar Flare Prediction Using Hybrid VGGNet-16 and LeNet for solar flare prediction, achieving
CNN- Model high accuracy on the SDO dataset and paving the way for
real-time alerts to enhance space weather forecasting.
 Methodology Model
Transformer, BiLSTM-Attention,
BiLSTM, LSTM-Attention, LSTM, and
neural network (NN)

Support Vector Machines

(SVMs),Multilayer Perceptrons
(MLPs),Decision Tree
Algorithms,Recurrent Neural
Networks (RNNs),Convolutional
Neural Networks (CNNs),Logistic
Regression, Ridge Classifier, Extreme
Gradient Boosting, Light Gradient
Boosting Machine, and Linear
Discriminant Analysis
 Deep learning , CNN(50 layers),
ResNet50 model
image processing techniques

VGGNet-16 for feature extraction

VGGNet-16 Model, LeNet Model
and LeNet for classification
 Which is effective Dataset
Transformer model stands out as Space-Weather HMI Activity Region Patches
the most effective in both (SHARP), high photospheric free energy
categorical and probabilistic density (HED) region data
prediction.Captures long-range
dependencies more effectively
than LSTM models.
 performance of HMI
magnetograms to other SDO AIA
datasets and found that HMI HMI magnetogram patch images from SDO,
magnetograms consistently ImageNet dataset ,SDOBenchmark dataset
yielded higher accuracy in
predicting extreme solar flares.

They not compared so, it’s

SDO Benchmark from Kaggle
unclear which is better alone.
Size of the image Classes of flares
 This dataset consists of 256x256 pixel
C, M, and X class flares.
images of solar flares

images to 224 x 224 pixels to fit the

VGGNet-16 model's, The dataset also
categorizes data into three image groups,
including images taken at the AIA
wavelength of 131 nanometers, Not Mentioned
magnetogram images from the Helioseismic
and Magnetic Imager (HMI), and
photographs recorded at the AIA
wavelength of 171 nanometers.
 Limitation of dataset Parameters
SHARP data may not capture finer details R VALUE parameter in the SHARP dataset,
preceding flare eruptions Efree (total photospheric free magnetic
energy density) parameter in the HED
dataset
 One significant limitation is the overall
scarcity of magnetogram training data,
-
which constrains the robustness and
generalizability of our model

The dataset focuses on images from specific

wavelengths:images from AIA wavelengths
of 131 and 171 nanometers, as well as Images from specific wavelengths:
magnetogram images from HMI, dataset AIA 131 nanometers, AIA 171 nanometers,
may not capture the full spectrum of solar Magnetogram images, Active Region ID,
activity, Limited Timeframe:the specific Date and Time
time period covered is not mentioned in the
excerpts provided
Time duration of images are taken Which method they have used to divide the dataset
10-fold cross-validation (CV) method into
training, validation, and testing sets, with a 60%,
20%, and 20% distribution, respectively
 -

Not mentioned Not mentioned

 Perfromance 1 Performance 2 Percentage
Out of 25 SHARP parameters, the For SHARP, the model's TSS SHARP (0.559 ± 0.087)
researchers selected 10 based on score typically ranges from performs around 55-65%
prior studies. From 11 HED 0.472 to 0.646, based on as well as random
parameters, 6 with weaker the average score of 0.559 guessing, while HED (0.721
correlations were chosen. with a variability of ±0.087. ± 0.084) performs 70-80%
This selection aimed to avoid better than random
redundancy due to strong predictions.
correlations between some
parameters.
 The study found that HMI
magnetograms, which
show the Sun's magnetic
performance of HMI fields, are better at
magnetograms to other predicting solar flares than
SDO AIA datasets and AIA images. HMI had
The test accuracy achieved using HMI
found that HMI 75.5% accuracy, while AIA
magnetograms to predict extreme
magnetograms had around 71%. This is
solar flares was 0.755.
consistently yielded higher because HMI focuses on
accuracy in predicting the Sun's magnetic
extreme solar flares. features, which are key for
flare prediction, while AIA
doesn't highlight these
details as much.

Training Accuracy: 96% ,

Validation Accuracy: 94% ,
Not mentioned Not mentioned Testing Accuracy: 92%
These acuuracy for
comined those models
 Summray
HED parameters are better than
SHARP parameters at predicting
solar flares. The Transformer
model performs very well,
especially with HED parameters,
showing higher TSS, HSS, and
BSS scores compared to SHARP.
The scores reflect how good the
model is at correctly predicting
flare classes, with HED being
more reliable than SHARP for
both categorical and
probabilistic forecasting.
 The study shows that HMI
magnetograms, focusing on the
Sun's magnetic fields, are more
effective for predicting solar
flares. Instead of using complex
derived parameters, the
researchers directly used
magnetic field patches,
simplifying the process and
highlighting the importance of
these magnetic features.

The paper combines VGGNet-16

and LeNet to predict solar flares,
achieving 96% training, 94%
validation, and 92% testing
accuracy using the SDO
Benchmark dataset, highlighting
the strength of hybrid deep
learning models.
1 Solar & Heliospheric Observatory (SOHO)

2 Solar Dynamics Observatory (SDO)

3 Advanced Composition Explorer (ACE)

4 Dataset for Solar Flare Prediction (DSFP)

5 Space-Weather Heliospheric and Magnetic Imager Active Region Patches (SHARPs)

6 Geostationary Operational Environmental Satellite (GOES)

7 High Photospheric Free energy Density (HED region)

The mission gathers data on the inner and outer structure of the Sun, including the
corona and solar winds. It has multiple remote sensing platforms and instruments
that measure composition, wavelengths of light, and energetic particles.

The focuses on understanding the Sun's influence on Earth by studying the solar
atmosphere. It's equipped with three remote sensing experiments (Atmospheric
Imaging Assembly, EUV Variability Experiment, and the Helioseismic and Magnetic
Imager).

The collects particles from various origins (solar, interplanetary, interstellar, and
intergalactic) and provides space weather reports and warnings. It uses instruments
like mass spectrometers, ionic charge analyzers, and magnetometers.

The combines records of magnetic attributes and solar flare data, utilizing sources
like the Joint Science Operations Center (JSOC), GOES, and SWPC archives.

The data provides information about active regions on the Sun.

X-ray flare The catalogs are used to identify and correlate flare events
The HED dataset contains solar parameters like solar flare classifications, magnetic
field data, X-ray flux, sunspot information, coronal mass ejections (CMEs), and space
weather metrics, all crucial for predicting solar flare activity and understanding its
impact on Earth.
 Dataset Full Name

SOHO Solar and Heliospheric Observatory

SDO Solar Dynamics Observatory

ACE Advanced Composition Explorer

NOVAA NOAA's Advanced Data Archive and Analysis

SHARP DSFP Solar HMI Active Region Patches / Data Set of Flares and Events

GOES Geostationary Operational Environmental Satellites

Solar Wind: The continuous flow of charged particles from the Sun, which impacts Earth's
magnetosphere.
Magnetic Field: Measurements of the Sun’s magnetic field, which plays a key role in solar
flares, coronal mass ejections, and sunspot activity.

X-ray Flux: A measure of the Sun’s X-ray emissions, often used to assess solar flare activity.
Energetic Particles: High-energy particles (protons, electrons, and ions) emitted by the Sun
during solar events such as flares and CMEs.
Solar Irradiance: The amount of solar energy reaching a particular area, important for
understanding solar variability and its impact on Earth.
Sunspots: Areas on the Sun’s surface that are cooler and have a higher magnetic field, often
associated with solar flares and CMEs.

Coronal Mass Ejections (CMEs): Large bursts of solar wind and magnetic fields rising above the
solar corona or being released into space, which can affect space weather on Earth.
 Parameters/Measurement

Solar wind, magnetic field, solar irradiance, coronal mass ejections (CMEs), solar activity (flares), and sunspot da

Solar surface temperature, magnetic field, solar irradiance, sunspot activity, coronal loops, solar flares, and promine

Solar wind particles (ions, electrons), energetic particles, solar wind density, speed, temperature, magnetic field parameters
energetic particle fluxes.

Solar flares, geomagnetic storms, solar wind, space weather impacts.

Solar active region magnetic fields, sunspot areas, flare events, and solar activity indices.

X-ray flux, proton flux, solar flare occurrence, energetic particles, geomagnetic indices, and cosmic ray observatio
 Description
SOHO provides detailed data on the Sun’s atmosphere and solar wind, including solar flares, coronal mass
ejections (CMEs), and the solar corona. It helps to understand space weather.
SDO provides high-resolution images of the Sun in multiple wavelengths (UV, visible, and X-ray), offering insights
into solar dynamics, including flare activity, sunspots, and magnetic fields.
ACE monitors solar wind conditions in real-time, providing crucial data for space weather forecasting and
understanding how solar wind affects Earth’s magnetosphere.
NOVAA archives NOAA's space weather data, including solar flare classifications, solar wind conditions, and
geomagnetic storm information.
SHARP (part of the Solar Dynamics Observatory) provides high-resolution data on the Sun’s active regions,
focusing on magnetic field configurations and flare events.
GOES provides real-time solar data, including X-ray flux from solar flares, solar energetic particles, and space
weather forecasts. GOES is also used for monitoring Earth’s space weather.
Year

1995

2010

1997