REDUKSI DIMENSI DATA GAIA-ESO SURVEY PDR 4 DENGAN PCA DAN AUTOENCODER
Rapid development of astronomical data in terms of its size and complexity creates many opportunities as well as new challenges to explore new knowledge from observations. However, the high dimensionality of astronomical data, such as spectra, makes it difficult to fully explore the information t...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Subjects: | |
| Online Access: | https://digilib.itb.ac.id/gdl/view/66257 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | Rapid development of astronomical data in terms of its size and complexity
creates many opportunities as well as new challenges to explore new knowledge
from observations. However, the high dimensionality of astronomical data,
such as spectra, makes it difficult to fully explore the information they contain.
Dimensionality reduction, as a part of machine learning techniques, is
an alternative to get important information from high dimensional data.
This study aims to investigate and apply two algorithms for dimension
reduction, namely Principal Component Analysis and Autoencoder, to try
and obtain radial velocity and effective temperature information extracted
from Gaia-ESO Survey Public Data Release 4. In addition, the effectiveness
of the two algorithms was also investigated.
The algorithm is applied to the standardized spectrum data without radial
velocity correction process, producing features in the final result related to the
standardization process. Regression was then carried out on several pairs of
reduced components that were deemed to be used as estimators of physical
parameters using Random Forest algorithm.
Application of the two algorithms shows that both of them can be used to
reduce and reconstruct the data used with the best error of 0.0396. Accuracy
was obtained above 77% and 93% respectively for the stars’ radial velocity and
effective temperature regression using the dimensionality reduction results |
|---|