SMOTE-NC dan Tomek Links pada Klasifikasi Pohon untuk Data Tidak Seimbang (Studi Kasus Status Penyakit Asma)

Wiradhiani, Auriga (2026) SMOTE-NC dan Tomek Links pada Klasifikasi Pohon untuk Data Tidak Seimbang (Studi Kasus Status Penyakit Asma). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221075-Undergraduate_thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Asma merupakan penyakit pernapasan kronis yang berkaitan dengan berbagai faktor komorbid. Pada data kesehatan, distribusi status penyakit asma sering tidak seimbang sehingga model klasifikasi cenderung lebih baik dalam mengenali kelas mayoritas. Penelitian ini bertujuan mengevaluasi pengaruh penerapan SMOTE-NC dan Tomek Links terhadap kinerja Classification and Regression Trees (CART) serta mengidentifikasi karakteristik status asma berdasarkan faktor komorbid. Data berasal dari National Health and Nutrition Examination Survey (NHANES) periode Agustus 2021–Agustus 2023 yang mencakup 1.955 penduduk Amerika Serikat berusia 20 tahun ke atas. Data dibagi menjadi data training dan testing dengan proporsi 70:30. Pembentukan pohon klasifikasi dilakukan melalui tiga skenario, yaitu menggunakan data training asli, data training yang diseimbangkan dengan SMOTE-NC, serta data training hasil SMOTE-NC yang dilanjutkan dengan pembersihan menggunakan Tomek Links. SMOTE-NC menghasilkan masing-masing 966 observasi pada kelas Asma dan Tidak Asma, sedangkan setelah penerapan Tomek Links tersisa masing-masing 869 observasi. Optimasi hyperparameter dilakukan menggunakan Grid Search dan Stratified 10-Fold Cross-Validation, dengan penanganan ketidakseimbangan kelas hanya diterapkan pada training fold. Hasil evaluasi menunjukkan bahwa skenario yang menggunakan SMOTE-NC dan Tomek Links menghasilkan balanced accuracy tertinggi sebesar 64,33%, dengan accuracy 62,80%, sensitivity 68,02%, specificity 60,63%, dan misclassification rate 37,20%. Skenario tersebut memberikan kemampuan yang lebih seimbang dalam mengklasifikasikan kelas Asma dan Tidak Asma. Riwayat alergi menjadi pemilah utama, sedangkan skor depresi memiliki tingkat kepentingan tertinggi, diikuti oleh BMI, riwayat penyakit paru obstruktif kronis, usia, dan jenis kelamin.
=======================================================================================================================================
Asthma is a chronic respiratory disease associated with various comorbid factors. In health data, the distribution of asthma-status classes is often imbalanced, causing classification models to perform better in identifying the majority class. This study aimed to evaluate the effects of applying the Synthetic Minority Over-sampling Technique for Nominal and Continuous Data (SMOTE-NC) and Tomek Links on the performance of Classification and Regression Trees (CART), as well as to identify the characteristics of asthma status based on comorbid factors. The data were obtained from the National Health and Nutrition Examination Survey (NHANES) conducted from August 2021 to August 2023 and included 1,955 United States residents aged 20 years and older. The data were divided into training and testing sets at a ratio of 70:30. Classification trees were developed under three scenarios: using the original training data, training data balanced using SMOTE-NC, and SMOTE-NC-balanced training data subsequently cleaned using Tomek Links. SMOTE-NC produced 966 observations in each of the Asthma and Non-Asthma classes, while 869 observations remained in each class after the application of Tomek Links. Hyperparameter optimization was performed using Grid Search and Stratified 10-Fold Cross-Validation, with class imbalance handling applied only to the training folds. The evaluation results showed that the scenario using SMOTE-NC and Tomek Links achieved the highest balanced accuracy of 64.33%, with an accuracy of 62.80%, sensitivity of 68.02%, specificity of 60.63%, and a misclassification rate of 37.20%. This scenario demonstrated a more balanced ability to classify the Asthma and Non-Asthma classes. A history of allergy was the primary splitting variable, while depression score had the highest variable importance, followed by BMI, history of chronic obstructive pulmonary disease, age, and sex.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Asma, CART, Komorbiditas, Tomek Links, Asthma, Comorbidity, SMOTE-NC
Subjects:	Q Science > QA Mathematics > QA76.9.D343 Data mining. Querying (Computer science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Auriga Wiradhiani
Date Deposited:	04 Aug 2026 04:54
Last Modified:	04 Aug 2026 04:54
URI:	http://repository.its.ac.id/id/eprint/142111

Actions (login required)

View Item