Implementasi Attention-based LSTM untuk Estimasi Angka Insiden Demam Berdarah Dengue di Kabupaten Malang Berbasis Explainable AI

Kharisma, Arjuna Putra (2026) Implementasi Attention-based LSTM untuk Estimasi Angka Insiden Demam Berdarah Dengue di Kabupaten Malang Berbasis Explainable AI. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026221210-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (11MB) | Request a copy

Abstract

Demam Berdarah Dengue (DBD) merupakan masalah kesehatan masyarakat yang signifikan di Kabupaten Malang, dengan pola kasus yang non-linear dan musiman yang kompleks. Penelitian ini bertujuan membangun model peramalan angka insiden DBD yang akurat sekaligus interpretatif menggunakan kombinasi Attention-based Long Short-Term Memory (Attention-LSTM), dekomposisi Classical Additive, dan SHapley Additive exPlanations (SHAP). Data 39 kecamatan periode Januari 2010–Juni 2023 diklasifikasikan ke dalam tiga klaster berdasarkan ketinggian dataran, dilatih dengan Huber Loss, dan divalidasi menggunakan Walk-Forward Validation. Model mencapai R² di atas 0,94 pada seluruh kecamatan representatif, dengan RMSE terbaik 0,4481 (Dataran Tinggi), 0,4192 (Dataran Sedang), dan 0,8427 (Dataran Rendah). Dekomposisi Classical Additive terbukti meningkatkan akurasi prediksi sebesar 78,8–89,2% dibandingkan baseline. Analisis SHAP mengidentifikasi variabel musiman bulan_sin dan bulan_cos sebagai faktor paling dominan di seluruh klaster, mengindikasikan bahwa siklus musiman tahunan menjadi penentu utama pola insiden DBD, dengan variabel iklim pendukung yang bervariasi antar klaster sesuai karakteristik geografisnya. Hasil penelitian membuktikan bahwa pendekatan hybrid Attention-LSTM dengan dekomposisi dan SHAP mampu menghasilkan peramalan DBD yang akurat dan dapat diinterpretasikan untuk mendukung mitigasi wabah.
================================================================================================================================
Dengue Hemorrhagic Fever (DHF) is a significant public health issue in Malang Regency, characterized by non-linear case patterns and complex seasonality. This study aims to build an accurate yet interpretable forecasting model for DHF incidence rates using a combination of Attention-based Long Short-Term Memory (Attention-LSTM), Classical Additive decomposition, and SHapley Additive exPlanations (SHAP). Data from 39 sub-districtsspanning January 2010 to June 2023 were classified into three clusters based on altitude, trained using Huber Loss, and validated via Walk-Forward Validation. The model achieved an R2 above 0.94 across all representative sub-districts, with the best RMSE values of 0.4481 (Highlands), 0.4192 (Midlands), and 0.8427 (Lowlands). Classical Additive decomposition proved to enhance forecasting accuracy by 78.8–89.2% compared to the baseline. Furthermore, SHAP analysis identified the seasonal variables bulan_sin and bulan_cos as the most dominant factors across all clusters, indicating that the annual seasonal cycle is the primary determinant of DHF incidence patterns, with supporting climate variables varying across clusters according to their geographical characteristics. The results demonstrate that the hybrid Attention-LSTM approach combined with decomposition and SHAP is capable of generating accurate and interpretable DHF forecasting to support outbreak mitigation.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Demam Berdarah Dengue, Peramalan Deret Waktu, Attention-based LSTM, Explainable Artificial Intelligence, SHapley Additive exPlanations, Kabupaten Malang, Dengue Hemorrhagic Fever, Time-Series Forecasting, Attention-based LSTM, Explainable Artificial Intelligence, SHapley Additive exPlanations, Malang Regency
Subjects:	T Technology > T Technology (General) > T174 Technological forecasting
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Arjuna Putra Kharisma
Date Deposited:	27 Jul 2026 00:52
Last Modified:	27 Jul 2026 00:52
URI:	http://repository.its.ac.id/id/eprint/137371

Available Versions of this Item

• Implementasi Attention-based LSTM untuk Estimasi Angka Insiden Demam Berdarah Dengue di Kabupaten Malang Berbasis Explainable AI. (deposited 27 Jul 2026 00:52) [Currently Displayed]
  • Implementasi Attention-based LSTM Untuk Estimasi Angka Insiden Demam Berdarah Dengue Di Kabupaten Malang Berbasis Explainable AI. (deposited 27 Jul 2026 01:40)

Actions (login required)

View Item