Claim Missing Document
Check
Articles

Found 2 Documents
Search

Performance Analysis of Tree-Based Models for Classifying Complete Basic Childhood Immunization in West Java Windi Pangesti; Mega Maulina; Hazelita Dwi Rahmasari; Bagus Sartono; Budi Susetyo; Aulia Rizki Firdawanti; Gerry Alfa Dito
Inferensi Vol 9 No 1 (2026)
Publisher : Department of Statistics ITS

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.12962/j27213862.v9i1.9167

Abstract

Complete basic immunization is a key public health indicator, and disparities in coverage remain a major concern in West Java. The 2024 Universal Child Immunization (UCI) rate in West Java reached only 77.47%, declining from 2023 and reflecting persistent disparities in access to immunization services, particularly between urban and rural areas. This study aims to identify the determinants of complete basic immunization among children in West Java using SUSENAS 2024 survey data (n = 4,672). Three tree- based classification algorithms CART, Random Forest, and LightGBM were applied, with class imbalance addressed using SMOTE, Tomek Links, and the SMOTE–Tomek Links hybrid method. Model performance was evaluated using balanced accuracy. The Random Forest model combined with SMOTE achieved the highest performance, with a balanced accuracy of 60.3% and an overall accuracy of 70.3%. This model also demonstrated superior capability in identifying children with incomplete immunization. Global feature importance results indicate that household spending category, KIA book ownership, maternal age at first birth, and maternal education are the strongest predictors of complete basic immunization. SHAP analysis reveals contrasting patterns: knowledge- based factors dominate in urban areas, while structural and socioeconomic constraints are more influential in rural areas. These findings underscore the importance of geographically targeted immunization strategies to support equitable access across urban and rural communities in West Java.
Performance Evaluation of Multinomial Logistic Regression, Random Forest, and XGBoost Methods in Data Classification Mega Maulina; Yani Prihantini Hiola; Indahwati; Aam Alamudi
Journal of Mathematics, Computations and Statistics Vol. 8 No. 2 (2025): Volume 08 Nomor 02 (Oktober 2025)
Publisher : Jurusan Matematika FMIPA UNM

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35580/jmathcos.v8i2.8459

Abstract

The development of data volume and complexity in the digital era increases the need for effective classification methods to support decision-making. Decision-making in classification tasks often requires methods that are well-suited to the data, along with the ability to produce accurate and reliable predictions. As scientific knowledge continues to advance, a wide range of classification methods have been developed. This study aims to analyze the performance of three commonly used classification methods Multinomial Logistic Regression, Random Forest, and XGBoost, in handling diverse data characteristics. Ten varied public datasets were used in this research, with differences in the number of classes, features, instances, balanced and imbalanced data conditions. Evaluation was conducted based on accuracy, F1-score, precision, and recall. The analysis results show that Random Forest consistently delivers the best performance particularly on imbalanced data. XGBoost demonstrates superiority on more complex datasets, while Multinomial Logistic Regression proves more effective on relatively small datasets. This research provides valuable insights into selecting appropriate classification methods based on data characteristics and highlights the effectiveness of ensemble-based approaches in handling diverse data. Based on the findings, it is recommended that the selection of classification algorithms be tailored to the characteristics of the dataset. Random Forest is preferable for handling imbalanced data, while XGBoost is ideal for complex datasets requiring robust hyperparameter tuning. Multinomial Logistic Regression remains a viable option for simpler datasets with fewer observations and features. Future research could explore hybrid models that combine these approaches to further optimize classification performance across various domains.