Hasanah, Silviatul (2018) Propensity Score Matching Menggunakan Support Vector Machine Pada Kasus Diabetes Melitus (DM) Tipe 2. Masters thesis, Institut Teknologi Sepuluh Nopember.
Preview |
Text
1316201034-Master_Thesis.pdf - Accepted Version Download (2MB) | Preview |
Abstract
Penelitian non-experimental dapat dilakukan diberbagai bidang salah satunya bidang kesehatan. Dalam bidang kesehatan Randomized Controlled Trials (RCT) tidak dapat dilakukan karena berhubungan dengan nyawa manusia. Pengacakan yang tidak dapat dilakukan pada penelitian non-experimental menyebabkan kovariat tidak seimbang antara kelompok treatment dan kontrol. Ketidakseimbangan ini menyebabkan estimasi efek perlakuan menjadi bias. Selain itu, adanya variabel confounding juga mengakibatkan estimasi efek perlakuan bias. Metode yang dapat digunakan untuk mengatasi estimasi yang bias pada efek perlakuan adalah metode Propensity Score (PS). Salah satu metode yang dikembangkan dari propensity score adalah metode Propensity Score Matching (PSM). Variabel confounding yang digunakan dalam penelitian ini adalah variabel aktivitas olahraga. Tujuan penelitian ini adalah mendapatkan hasil estimasi propensity score menggunakan metode support vector machine dan regresi logistik biner, kemudian setelah mendapatkan nilai estimasi tersebut menghitung hasil bias sebelum dilakukan matching serta bias setelah dilakukan matching dengan menggunakan metode propensity score matching support vector machine dan propensity score matching regresi logistik biner. Selanjutnya mendapatkan signifikansi hasil Average Treatment of Treated (ATT) untuk melihat pengaruh antara variabel confounding (Z) dengan variabel respon (Y). Data yang digunakan dalam penelitian ini adalah data pasien Diabetes Melitus (DM) tipe 2 yang di rawat di Rumah Sakit Umum Daerah (RSUD) Pasuruan pada periode Maret 2016. Signifikansi hasil ATT untuk PSM-SVM maupun PSM-regresi logistik biner menunjukkan bahwa variabel aktivitas olahraga (Z) berpengaruh secara signifikan terhadap variabel komplikasi penyakit (Y). Metode PSM-SVM mampu mereduksi bias sebesar 16,65% dan metode PSM-regresi logistik biner mampu mereduksi bias sebesar 43,76%. Metode propensity score matching support vector machine merupakan metode yang lebih baik dibandingkan metode propensity score matching regresi logistik biner karena menghasilkan nilai bias setelah matching yang lebih kecil yaitu sebesar 0,0836.
====================================================================== Non-experimental study can be done in various fields one of them is health field. In the health field Randomized Controlled Trials (RCT) can not be done because it related to human life. Randomization that can not be done in non-experimental studies causes unbalanced covariates between treatment and control groups. This imbalance causes the estimation of the treatment effect was biased. In addition, the presence of confounding variables also results in biased estimation of treatment effect. The appropriate method to handle the bias estimation of treatment effect is Propensity Score (PS) method. One of the methods that have been developed from the propensity score is Propensity Score Matching (PSM). Confounding variables that used in this study is exercise activities. The purpose of this study is to get the result of propensity score estimation using support vector machine method and binary logistic regression, then after getting the estimated value, calculate bias before matching and bias after matching by using propensity score matching-support vector machine and propensity score matching-binary logistic regression. Furthermore getting the significance of Average Treatment of Treated (ATT) to see the effect of confounding variable (Z) with response variable (Y). The data used in this study is type 2 Diabetes Mellitus (DM) patients data treated at Pasuruan regional public hospital on March period 2016. Significance results of Average Treatment of Treated (ATT) for PSM-SVM and PSM-binary logistic regression showed that the exercise activity variables (Z) has significant influence for disease complication variables (Y). The PSM-SVM method can reduce the bias by 16.65% and the binary logistic PSM-regression method can reduce the bias by 43.76%. The PSM-support vector machine method is a better than PSM-binary logistic regression method because it produces a bias value after matching smaller, namely 0.083.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | ATT, Bias, Confounding, Diabetes Melitus, Non-experimental, Propensity Score Matching, Regresi Logistik Biner, SVM |
| Subjects: | Q Science > Q Science (General) Q Science > Q Science (General) > Q180.55.M38 Mathematical models Q Science > Q Science (General) > Q325 GMDH algorithms. R Medicine > R Medicine (General) |
| Divisions: | Faculty of Mathematics, Computation, and Data Science > Statistics > 49101-(S2) Master Thesis |
| Depositing User: | Silviatul Hasanah |
| Date Deposited: | 18 Jun 2021 11:28 |
| Last Modified: | 18 Jun 2021 11:28 |
| URI: | http://repository.its.ac.id/id/eprint/55604 |
Actions (login required)
 |
View Item |