Electronic Theses and Dissertation

Prognosis setelah stroke hemoragik sering kali sulit ditentukan, terutama dalam 24–48
jam pertama ketika pengambilan keputusan klinis sangat krusial. Prediksi mortalitas
dini berperan penting dalam proses triase dan perencanaan terapi. Meskipun
pendekatan machine learning telah banyak diterapkan dalam penelitian mengenai
stroke, studi yang secara khusus mengembangkan model prediktif awal untuk
mortalitas pada stroke hemoragik akut masih terbatas, terutama di Indonesia. Penelitian
ini bertujuan untuk merancang model machine learning yang dapat memprediksi
mortalitas dini pada pasien stroke hemoragik akut menggunakan data retrospektif
pasien dengan perdarahan intraserebral yang dirawat di rumah sakit rujukan. Dataset
mencakup 746 pasien dengan 20 variabel klinis, laboratorium, dan radiologis, serta
menggunakan tiga algoritma pembanding, yaitu Decision Tree, Random Forest, dan
Gaussian Naive Bayes. Kinerja model dievaluasi berdasarkan akurasi, sensitivitas,
spesifisitas, F1-score, dan Area Under the Curve (AUC). Hasil menunjukkan bahwa
Random Forest memberikan performa terbaik dengan akurasi 84,77%, F1-score
84,63%, dan AUC 80,51%, diikuti oleh Decision Tree dengan akurasi 80,98% dan
AUC 61,12%, serta Naive Bayes dengan akurasi 68,35%, presisi tertinggi 89,83%, dan
AUC 82,94%. Model Random Forest terbukti paling unggul dalam memberikan
prediksi dini terhadap mortalitas pasien stroke hemoragik dengan akurasi dan stabilitas
yang lebih tinggi dibandingkan algoritma lainnya. Implementasi model ini berpotensi
membantu klinisi dalam proses triase, pengambilan keputusan, dan penatalaksanaan
pasien berisiko tinggi, meskipun penelitian lanjutan dengan data multi-senter dan
variabel yang lebih komprehensif masih diperlukan untuk meningkatkan validitas dan
generalisasi model.
Kata kunci: Stroke hemoragik, ICH, mortalitas, prediksi dini, machine learning.

Prognosis after hemorrhagic stroke is often challenging to establish, particularly within the first 24–48 hours when clinical decision-making is critical. Early mortality prediction plays an essential role in triage and treatment planning. Although machine learning has been widely applied in stroke studies, studies specifically developing early predictive models for mortality in acute hemorrhagic stroke remain limited, especially in Indonesia. This study aimed to design a machine learning model to predict early mortality in patients with acute hemorrhagic stroke. A retrospective dataset of patients with intracerebral hemorrhage admitted to a referral hospital was analyzed. The dataset consisted of 746 patients with 20 clinical, laboratory, and radiological variables. The algorithms compared are the decision tree, the random forest, and the Gaussian Naive Bayes. Model performance was evaluated using accuracy, sensitivity, specificity, F1-score, and area under the curve (AUC). The results showed that random forest achieved the best performance, with an accuracy of 84.77%, F1-score of 84.63%, and AUC of 80.51%. The decision tree reached an accuracy of 80.98% but with a low AUC (61.12%). Naive Bayes recorded the lowest accuracy (68.35%) despite having the highest precision (89.83%) and an AUC of 82.94%. In conclusion, machine learning models, particularly random forest, demonstrated the ability to provide early mortality prediction in hemorrhagic stroke patients with better accuracy and stability compared to other algorithms. The implementation of such models has the potential to assist clinicians in triage, decision-making, and management of high-risk patients. However, further research with multi-center datasets and more comprehensive variables is required to enhance validity and generalizability. Keywords: Hemorrhagic stroke, intracerebral hemorrhage (ICH), mortality, early prediction, machine learning.