Skip to main navigation Skip to search Skip to main content

Evaluating Chi-Square-Based Discretization and Ensemble Learning for Patient Care Classification Using Structured Electronic Health Records Data

  • Institut Teknologi Sepuluh Nopember

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Accurate classification of patient treatment status (outpatient and inpatient) from Electronic Health Records (EHR) is essential for timely clinical decision-making and efficient resource allocation. However, many studies still rely on single classifiers and report only moderate performance while the practical contribution of discretization and ensemble learning remains underexplored. This study proposes an integrated evaluation framework that systematically evaluates combinations of chi-square-based discretization methods, base classifiers, and ensemble strategies for patient treatment status classification. The framework is evaluated on a public EHR dataset from private hospitals in Indonesia, considering four chi-square discretization algorithms (ChiMerge, Chi2, ExtChi2, and ModChi2), three optimized base learners (Random Forest, XGBoost, and MLP), and three ensemble approaches (hard voting, optimized soft voting, and stacking). To ensure fair comparisons, hyperparameters are tuned using Bayesian optimization (Optuna) under a 5-fold cross-validation. Results show that chi-square discretization provides limited benefit; the best single-model performance is achieved by XGBoost without discretization (F1-score 76.64%, AUC-ROC 81.78%). The best overall result is obtained by an optimized soft-voting ensemble of XGBoost and MLP, reaching 77.02% F1-score with a comparable 81.75% AUC-ROC and more stable predictions across folds than individual models. Overall, for relatively clean and structured EHR data, well-tuned strong learners and lightweight ensembles are more impactful than advanced discretization techniques.

Original languageEnglish
Title of host publication27th International Seminar on Intelligent Technology and Its Applications, ISITIA 2026
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages874-879
Number of pages6
ISBN (Electronic)9798319521576
DOIs
Publication statusPublished - 2026
Event27th International Seminar on Intelligent Technology and Its Applications, ISITIA 2026 - Hybrid, Bali, Indonesia
Duration: 22 Jul 2026 → 24 Jul 2026

Conference

Conference27th International Seminar on Intelligent Technology and Its Applications, ISITIA 2026
Country/TerritoryIndonesia
CityHybrid, Bali
Period22/07/26 → 24/07/26

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

  1. SDG 7 - Affordable and Clean Energy
    SDG 7 Affordable and Clean Energy

Keywords

  • Chi-square Discretization
  • Classification
  • Electronic Health Records
  • Ensemble Learning
  • Soft Voting

Fingerprint

Dive into the research topics of 'Evaluating Chi-Square-Based Discretization and Ensemble Learning for Patient Care Classification Using Structured Electronic Health Records Data'. Together they form a unique fingerprint.

Cite this