Evaluation performance of Hybrid Localized Multi Kernel SVR (LMKSVR) in electrical load data using 4 different optimizations

Rezzy Eko Caraka, Sakhinah Abu Bakar

Research output: Contribution to journalArticle

6 Citations (Scopus)

Abstract

The main problem using SVR is to find optimal parameter (·) by using kernel function such as radial basis, polynomial, Gaussian and so on. Moreover, we also have to find optimal hyperplane parameter (C and ·). In the heart of statistical methods and data mining, the motivation of researcher doing this is to minimize time, money and energy in the analysis at the same time the results will be more accurate. The development of such a massive technology and the availability of data is very much making progress and improvement of methods based on data mining and machine learning. In this paper, we proposed four different optimizations such as LIBSVM, MOSEK, QUADPROG, SMO applied to Localized Multi-Kernel Learning (LMKL) to assign local weights to kernel functions so that the best hyperplane parameters will be obtained. For the simulation, we use the electrical data, and we have labeled based on the characteristics of different days (Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, Saturday, National Holiday, Ramadhan). As well as, we can capture the pattern of electricity consumption.

Original languageEnglish
Pages (from-to)7440-7449
Number of pages10
JournalJournal of Engineering and Applied Sciences
Volume13
Issue number17
Publication statusPublished - 1 Jan 2018

Fingerprint

Data mining
Learning systems
Statistical methods
Electricity
Polynomials
Availability

Keywords

  • Electrical
  • Hybrid
  • Kernel
  • Optimization
  • SVR

ASJC Scopus subject areas

  • Engineering(all)

Cite this

@article{ea3f11ddb2a0477d9671a042f231ff1b,
title = "Evaluation performance of Hybrid Localized Multi Kernel SVR (LMKSVR) in electrical load data using 4 different optimizations",
abstract = "The main problem using SVR is to find optimal parameter (·) by using kernel function such as radial basis, polynomial, Gaussian and so on. Moreover, we also have to find optimal hyperplane parameter (C and ·). In the heart of statistical methods and data mining, the motivation of researcher doing this is to minimize time, money and energy in the analysis at the same time the results will be more accurate. The development of such a massive technology and the availability of data is very much making progress and improvement of methods based on data mining and machine learning. In this paper, we proposed four different optimizations such as LIBSVM, MOSEK, QUADPROG, SMO applied to Localized Multi-Kernel Learning (LMKL) to assign local weights to kernel functions so that the best hyperplane parameters will be obtained. For the simulation, we use the electrical data, and we have labeled based on the characteristics of different days (Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, Saturday, National Holiday, Ramadhan). As well as, we can capture the pattern of electricity consumption.",
keywords = "Electrical, Hybrid, Kernel, Optimization, SVR",
author = "Caraka, {Rezzy Eko} and {Abu Bakar}, Sakhinah",
year = "2018",
month = "1",
day = "1",
language = "English",
volume = "13",
pages = "7440--7449",
journal = "Journal of Engineering and Applied Sciences",
issn = "1816-949X",
publisher = "Medwell Journals",
number = "17",

}

TY - JOUR

T1 - Evaluation performance of Hybrid Localized Multi Kernel SVR (LMKSVR) in electrical load data using 4 different optimizations

AU - Caraka, Rezzy Eko

AU - Abu Bakar, Sakhinah

PY - 2018/1/1

Y1 - 2018/1/1

N2 - The main problem using SVR is to find optimal parameter (·) by using kernel function such as radial basis, polynomial, Gaussian and so on. Moreover, we also have to find optimal hyperplane parameter (C and ·). In the heart of statistical methods and data mining, the motivation of researcher doing this is to minimize time, money and energy in the analysis at the same time the results will be more accurate. The development of such a massive technology and the availability of data is very much making progress and improvement of methods based on data mining and machine learning. In this paper, we proposed four different optimizations such as LIBSVM, MOSEK, QUADPROG, SMO applied to Localized Multi-Kernel Learning (LMKL) to assign local weights to kernel functions so that the best hyperplane parameters will be obtained. For the simulation, we use the electrical data, and we have labeled based on the characteristics of different days (Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, Saturday, National Holiday, Ramadhan). As well as, we can capture the pattern of electricity consumption.

AB - The main problem using SVR is to find optimal parameter (·) by using kernel function such as radial basis, polynomial, Gaussian and so on. Moreover, we also have to find optimal hyperplane parameter (C and ·). In the heart of statistical methods and data mining, the motivation of researcher doing this is to minimize time, money and energy in the analysis at the same time the results will be more accurate. The development of such a massive technology and the availability of data is very much making progress and improvement of methods based on data mining and machine learning. In this paper, we proposed four different optimizations such as LIBSVM, MOSEK, QUADPROG, SMO applied to Localized Multi-Kernel Learning (LMKL) to assign local weights to kernel functions so that the best hyperplane parameters will be obtained. For the simulation, we use the electrical data, and we have labeled based on the characteristics of different days (Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, Saturday, National Holiday, Ramadhan). As well as, we can capture the pattern of electricity consumption.

KW - Electrical

KW - Hybrid

KW - Kernel

KW - Optimization

KW - SVR

UR - http://www.scopus.com/inward/record.url?scp=85055514819&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85055514819&partnerID=8YFLogxK

M3 - Article

VL - 13

SP - 7440

EP - 7449

JO - Journal of Engineering and Applied Sciences

JF - Journal of Engineering and Applied Sciences

SN - 1816-949X

IS - 17

ER -