Prognosis pada Pasien Diabetes Mellitus Tipe 2 Terhadap Penyakit Komplikasinya dengan Metode Lifelong Learning Decision Tree Classifier
Anik Andriani, Prof. Dra. Sri Hartati, M.Sc., Ph.D; Afiahayati, S.Kom., M.Cs., Ph.D
2025 | Disertasi | S3 Ilmu Komputer
The prognosis of patients with Type 2 Diabetes Mellitus (T2DM), regarding the risk of complications and the likelihood of control, is necessary to determine appropriate treatment recommendations. Challenges in implementing classification in this case include the limited availability of complete T2DM patient data across the seven categories of complications and differences in patient medical record data formats from one hospital to another. Data on T2DM patients collected from 2017 to 2022 at Bethesda Hospital only provided five complication categories. Meanwhile, data on T2DM patients collected from the same years at Panti Rapih Hospital provided seven complication categories. The classification results from the Panti Rapih Hospital dataset using conventional machine learning methods cannot be directly applied to data from other hospitals. This is because there are often differences in data formats and the amount of data in patient medical records, resulting in varying feature formats and the number of features in the dataset.
This dissertation research proposes a Lifelong Learning classification model with a Decision Tree Classifier. The Decision Tree method was developed within the Lifelong Learning classification model because it provides the best performance for classification in the source domain. The medical records of T2DM patients from Panti Rapih Hospital served as the source domain due to its more comprehensive data. The Transfer Learning method was applied to transfer knowledge from the source domain, including source domain features, feature weights, and classification rules, to the target domain. This knowledge transfer was used to build a knowledge transfer model within the Lifelong Learning Decision Tree classification model. Furthermore, the knowledge-based learner developed domain adaptation capabilities based on a knowledge base and task-specific generator containing new data from various target domains with varying feature formats. The goal was to test the Lifelong Learning model's ability to adapt to new features and address missing features.
The application of the Lifelong Learning Decision Tree classification model to 16 target domains with varying feature formats demonstrated a better average accuracy value than the conventional Decision Tree method. The average accuracy of the proposed model for classification in the target domain was 0.895, while the average accuracy of classification results in the target domain using the conventional Decision Tree method was 0.696. The performance of the proposed model, calculated using Average Accuracy, showed an accuracy value of 0.874.
Kata Kunci : Klasifikasi, Lifelong Learning, Prognosis Diabetes Mellitus Tipe 2, Decision Tree