Claim Missing Document
Check
Articles

Oversampling Techniques to Improve Machine Learning Classification on a Heart Disease Dataset Rohaini, Eni; Gunardi; Nurhayati; Jasmir; Prisdian Tiararosa, Zahra
Jurnal Manajemen Teknologi Dan Sistem Informasi (JMS) Vol 6 No 2 (2026): JMS Vol 6 No 2 September 2026
Publisher : LPPM Universitas Dinamika Bangsa

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33998/jms.2026.6.2.2842

Abstract

Class imbalance in heart disease datasets poses a critical challenge in machine learning applications, as models often become biased toward the majority class. This imbalance reduces accuracy and limits the model’s ability to identify minority cases, which are highly important in medical diagnosis. This study investigates the effectiveness of oversampling techniques, namely Random Oversampling and Synthetic Minority Over-sampling Technique (SMOTE), in enhancing the performance of three classification algorithms: K-Nearest Neighbors (KNN), Naive Bayes (NB), and Random Forest (RF). The dataset, obtained from Kaggle, consists of 918 records with 12 attributes representing patient health indicators related to heart disease prediction. The research process involved data preprocessing, baseline classification experiments, and re-experiments using the oversampling methods. The results show that all three algorithms benefited from oversampling, though the degree of improvement varied. KNN achieved its best performance with SMOTE, reaching 72.98% accuracy and an F1-score of 75.39%. Naive Bayes produced relatively stable results across both methods, with SMOTE slightly outperforming Random Oversampling, achieving the best F1-score of 73.56%. Random Forest demonstrated the highest overall performance when combined with Random Oversampling, recording 79.19% accuracy and an F1-score of 81.51%, surpassing both the baseline and SMOTE. These findings confirm that oversampling techniques can effectively improve classification performance but their impact depends on the characteristics of each algorithm. The novelty of this study lies in analyzing the interaction between oversampling strategies and classification models, offering practical guidance for selecting appropriate data balancing approaches in medical classification tasks
Comparison of CRF and BiLSTM Models for Disease Named Entity Recognition on the NCBI Disease Dataset Agus Siswanto; Bambang Tutuko; Firdaus; Jasmir
Computer Engineering and Applications Journal Vol. 15 No. 3 (2026)
Publisher : Universitas Sriwijaya

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.18495/comengapp.v15i3.1352

Abstract

Named Entity Recognition (NER) is a fundamental task in Natural Language Processing (NLP) aimed at identifying and classifying named entities in text. In the biomedical domain, disease entity recognition is crucial for supporting automated medical information extraction. This study compares two NER model approaches, Conditional Random Field (CRF) and Bidirectional Long Short-Term Memory (BiLSTM), for disease entity recognition using the NCBI Disease dataset. The CRF model utilizes handcrafted linguistic features including prefixes, suffixes, POS tags, and word shapes, while BiLSTM employs GloVe 200-dimensional embeddings and automatically learns features from sequential context. BiLSTM experiments were conducted using 5 different seeds for robust statistical validation, reporting mean F1-score, standard deviation, and 95% confidence interval. Comprehensive analysis includes confusion matrix, error analysis, learning curve, and performance comparison per entity length. Results show that CRF achieves an F1-score of 78.49% (P=82.37%, R=74.95%), while BiLSTM achieves 76.54% ± 1.32% (95% CI ±1.15%). CRF outperforms BiLSTM by 1.95 percentage points under baseline conditions, although BiLSTM demonstrates notably fewer false positives (FP=124 vs FP=153) and a standard deviation of 1.32% across 5 experimental seeds