Stellar Signal Feature Priority with Differential Evolution

• Published in: 2018 Seventeenth Mexican International Conference on Artificial Intelligence (MICAI) pp. 76 - 80
• Main Authors: Contreras Sanchez, Daniel, Flores Pulido, Leticia, Portilla Flores, Edgar Alfredo, Calva Yanez, Maria Barbara
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-10-2018
• Subjects: bio-inspired algorithms; differential evolution; evolutive learning; feature selection; stellar items
• DOI: 10.1109/MICAI46078.2018.00020

Stellar classification inside astrophysics, is not a trivial task, because huge data amount of stellar objects in the sky. Telescopes in earth or in atmosphere allows to provide this data amount about stellar bodies. Data availability is overwhelming in some cases and automatized processes can provide a fast classification in stellar data despite data redundancy and to improve data relevance in data sets. Nature of stellar data, some features may be irrelevant in pulsar classification, star, or constellation, galaxy cluster or nebula. Stellar samples must have a noise elimination. Electromagnetic spectre must be considered in order to obtain samples of telescopes. Data relevance is affected by hertz value range of observation. Weather conditions, technical specifications in telescopes and data redundancy have one strong influence in data samples. Every feature observed has a numerical range that are considered as part of the restrictions of the evolutive algorithm considered in this research. Priority about features observed, can imply a faster classification of data. However, restrictions must be considered in order to stablish relations between other features and finally to state an objective function. The optimization problem in this case, involve 40,000 variations of relations among data, solutions, variables, type of star and restrictions. The data set is of 740 stars with 7 features. This research proposes a feature priority with differential evolution. The priority of features may be helpful those researches who desire a classification faster and efficient, despite data redundancy or singular values in a classification future phase in stellar data.