Maghfiroh, Aulia (2013) Perbandingan Metode Partial Least Absolute Deviation Regression Dan Metode Partial Robust M Regression Pada Kasus Data Outlier. Sarjana thesis, Universitas Brawijaya.
Abstract
Partial Least Square Regression (Regresi PLS) merupakan metode yang sangat berguna untuk memprediksi variabel respon berdasarkan variabel prediktor berdimensi tinggi (banyak pengamatan lebih sedikit dibanding banyak variabel prediktor). Namun regresi PLS sangat rentan dengan adanya data outlier. Outlier pada regresi PLS di antaranya leverage outlier, vertical outlier dan orthogonal outlier. Pendekatan robust dari regresi PLS yang dapat mengatasi data outlier adalah metode Partial Least Absolute Deviation Regression (Regresi PLAD) dan Partial Robust M Regression (Regresi PRM). Tujuan dari penelitian ini untuk membandingkan kebaikan metode regresi PLAD dan regresi PRM dalam mengatasi data outlier. Data yang digunakan berupa 3 data simulasi yang masing-masing dibangun mengandung 3 jenis outlier yang berbeda serta 1 data simulasi yang tidak mengandung outlier. Indikator kebaikan metode dan pemilihan metode terbaik adalah berdasarkan Root Mean Square Error (RMSE) serta Goodness of Fit (GOF). Hasil penelitian ini menunjukkan bahwa berdasar indikator kebaikan GOF maupun RMSE pada saat data tidak mengandung outlier didapatkan bahwa metode regresi PLAD memberikan hasil pendugaan yang lebih baik dibanding metode regresi PRM namun metode regresi PLS paling baik. Pada data yang masing-masing mengandung 10% leverage outlier dan 10% orthogonal outlier, nilai GOF dan RMSE untuk metode regresi PLAD dan regresi PRM hampir sama sehingga kedua metode terbukti sama baik berdasarkan kriteria GOF dan RMSE. Metode regresi PLAD lebih baik dalam memodelkan data yang mengandung 10% vertical outlier dibanding metode regresi PRM.
English Abstract
Partial Least Square Regression (PLS) is a useful method for predict respon variable from high dimensional predictor variable (number of observation less than number of predictor). However, PLS Regression method is highly influence with outlier data. There are three kinds of outlier in PLS regression, such as leverage outlier, vertical outlier and orthogonal outlier. Robustified version of PLS Regression which can deal with outlier data are Partial Least Absolute Deviation Regression (PLAD) and Partial Robust M Regression (PRM). The objective study of this research is to compare how well PLAD Regression and PRM Regression fitted the data contains outlier. Four data will construct in this research, which three of them are data with different tipe of outlier and other is data with no outlier. The goodness of these method is evaluated by Root Mean Square Error (RMSE) and Goodness of Fit (GOF) statistics. This statistics is also used as the selection of the best method. The result of this research show that when data have no outlier, the PLS Regression method perform better than its robust versions. PLAD Regression have better GOF and RMSE than PRM Regression for this data. On each data containing 10% leverage outlier and 10% orthogonal outlier, PLAD Regression and PRM regression gives GOF and RMSE results are almost the same. Therefore, both methods provide similar estimates for those data type. When 10% vertical outlier is added, PLAD regression give better fitted than two others methods
| Item Type: | Thesis (Sarjana) |
|---|---|
| Identification Number: | SKR/MIPA/2013/148/051306321 |
| Subjects: | 500 Natural sciences and mathematics > 519 Probabilities and applied mathematics > 519.5 Statistical mathematics |
| Divisions: | Fakultas Matematika dan Ilmu Pengetahuan Alam > Statistika |
| Depositing User: | Hasbi |
| Date Deposited: | 02 Sep 2013 10:12 |
| Last Modified: | 25 Oct 2021 01:54 |
| URI: | http://repository.ub.ac.id/id/eprint/153407 |
Preview |
Text
Lampiran.pdf Download (5MB) | Preview |
Preview |
Text
Halaman_Depan.pdf Download (1MB) | Preview |
Preview |
Text
BAB_I-V.pdf Download (3MB) | Preview |
Preview |
Text
COVER.pdf Download (1MB) | Preview |
Actions (login required)
 |
View Item |