Pemodelan Relasi antara Korban dan Pelaku pada Kasus Kekerasan terhadap Perempuan dan Anak Menggunakan Bayesian Bernoulli Mixture Regression Model

Hendriani, Revana Moza (2026) Pemodelan Relasi antara Korban dan Pelaku pada Kasus Kekerasan terhadap Perempuan dan Anak Menggunakan Bayesian Bernoulli Mixture Regression Model. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221009_Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Fenomena kekerasan terhadap perempuan dan anak mencerminkan kompleksitas sosial yang ditandai oleh heterogenitas karakteristik korban serta dinamika relasi antara korban dan pelaku. Penelitian ini bertujuan mengidentifikasi karakteristik kasus, mengestimasi pengaruh variabel sosiodemografi terhadap jenis relasi antara pelaku dan korban, serta membandingkan performa Bernoulli Mixture Regression Model (BMRM) Bayesian terhadap Binary Logistic Regression Model Bayesian. Data administratif laporan kasus dimodelkan dengan variabel respons jenis relasi (personal dan nonpersonal) serta kovariat yang mencakup kategori, status perkawinan, tingkat pendidikan, dan kepemilikan jaminan kesehatan korban. Pemodelan BMRM Bayesian diestimasi menggunakan algoritma No-U-Turn Sampler (NUTS) dengan mengintegrasikan dua komponen struktur laten berbasis lokasi kejadian (privat dan publik), yang secara komputasi terbukti memenuhi kriteria konvergensi. Hasil inferensi memperlihatkan keberadaan heterogenitas struktur, yang ditunjukkan oleh variasi signifikansi kovariat antar-komponen. Pada komponen privat, pengaruh signifikan didominasi oleh kategori dan pendidikan korban, sedangkan pada komponen publik, seluruh kovariat terbukti berpengaruh signifikan. Evaluasi performa melalui Pareto Smoothed Importance Sampling Leave-One-Out Cross-Validation (PSIS-LOO) mengonfirmasi keunggulan BMRM Bayesian dengan nilai Leave-One-Out Information Criterion (LOOIC) sebesar 1.128,93, jauh melampaui model tunggal dengan nilai LOOIC sebesar 1.514,43. Eksplorasi prediktor linear (η) lebih lanjut mengungkap bahwa observasi pada kuartil ekstrem mencerminkan karakteristik yang paling representatif pada masing-masing komponen, sementara observasi di sekitar batas keputusan menggambarkan wilayah transisi antar-komponen akibat dinamika pelaporan dan struktur sosial yang kompleks. Dengan demikian, BMRM Bayesian tidak hanya unggul secara prediktif, tetapi juga mampu memberikan pemahaman empiris yang lebih komprehensif mengenai struktur heterogenitas data.
======================================================================================================================================
The phenomenon of violence against women and children reflects social complexity characterized by the heterogeneity of victim characteristics and the relational dynamics between victims and perpetrators. This study aims to identify case characteristics, estimate the effects of sociodemographic variables on the type of victim–perpetrator relationship, and compare the performance of the Bayesian Bernoulli Mixture Regression Model (BMRM) against the Bayesian Binary Logistic Regression Model. Administrative case report data were modeled using relationship type (personal and non-personal) as the response variable, alongside covariates including victim category, marital status, education level, and health insurance ownership. The Bayesian BMRM was estimated using the No-U-Turn Sampler (NUTS) algorithm by incorporating two latent structure components based on incident location (private and public), which computationally met convergence criteria. Inference results demonstrated structural heterogeneity, evidenced by variation in covariate significance across components. In the private component, significant effects were driven primarily by victim category and education, whereas in the public component, all covariates were statistically significant. Model evaluation via Pareto Smoothed Importance Sampling Leave-One-Out Cross-Validation (PSIS-LOO) confirmed the superiority of the Bayesian BMRM, yielding a Leave-One-Out Information Criterion (LOOIC) value of 1,128.93, outperforming the single-component model (1,514.43). Further exploration of the linear predictor (η) revealed that observations in the extreme quartiles represent the most characteristic profiles of their respective components, whereas observations near the decision boundary capture a transition zone between components driven by reporting dynamics and complex social structures. Consequently, the Bayesian BMRM not only demonstrates predictive superiority but also offers a more comprehensive empirical understanding of data heterogeneity structures.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Bernoulli Mixture Regression Model, Binary Logistic Rgression Model, Inferensi Bayesian, Kasus Kekerasan terhadap Perempuan dan Anak, Relasi Korban dan Pelaku
Subjects:	H Social Sciences > HA Statistics > HA31.3 Regression. Correlation. Logistic regression analysis. Q Science > QA Mathematics > QA278.2 Regression Analysis. Logistic regression Q Science > QA Mathematics > QA279.5 Bayesian statistical decision theory.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Revana Moza Hendriani
Date Deposited:	05 Aug 2026 08:03
Last Modified:	05 Aug 2026 08:03
URI:	http://repository.its.ac.id/id/eprint/144065

Actions (login required)

View Item