Deteksi Anomali Pada Data Time Series Konsentrasi PM2.5 Dari Stasiun Pemantau Kualitas Udara Kementerian Lingkungan Hidup Menggunakan REDLAMP

Athaillah, Alendra Rafif (2026) Deteksi Anomali Pada Data Time Series Konsentrasi PM2.5 Dari Stasiun Pemantau Kualitas Udara Kementerian Lingkungan Hidup Menggunakan REDLAMP. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221297-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Pada data PM2.5 dari SPKU, deteksi anomali time-series diperlukan untuk menandai titik atau subsekuens yang menyimpang dari pola normal sebelum data digunakan dalam analisis lebih lanjut, dan karena data yang digunakan belum tentu menyediakan label anomali, diperlukan pendekatan unsupervised yang mampu mendeteksi penyimpangan berdasarkan pola yang dipelajari dari data. Pendekatan unsupervised yang telah dikembangkan masih memiliki keterbatasan dalam menangani kontaminasi anomali, diversity gap, false anomalies, dan overconfidence pada label. Untuk mengatasi permasalahan tersebut, penelitian ini menerapkan REDLAMP (Robust and Explainable Detector of Time-Series Anomaly), yaitu metode unsupervised yang menggabungkan reconstruction dan classification dengan pendekatan multiclass, untuk mendeteksi anomali pada data time-series konsentrasi PM2.5 dari SPKU Gelora Bung Karno milik Kementerian Lingkungan Hidup. Melalui kombinasi reconstruction error dan adjusted anomaly-class score, model menghasilkan skor anomali yang menunjukkan tingkat penyimpangan suatu window terhadap pola yang dipelajari, dan kinerja REDLAMP dievaluasi berdasarkan variasi ukuran window dan jenis pseudo-anomali hasil augmentasi. Hasil evaluasi menunjukkan bahwa REDLAMP dapat membedakan titik anomali dan titik normal, dengan VUS-ROC tertinggi sebesar 0,8489 dan VUS-PR tertinggi sebesar 0,1800. Komponen reconstruction only secara konsisten menghasilkan performa tertinggi dibandingkan combined dan classification only, window yang lebih panjang menurunkan kemampuan model dalam memberi ranking anomali secara global, dan konfigurasi enam kelas pseudo-anomali (K=6) memperoleh performa tertinggi dibandingkan K=2 maupun K=12. Secara keseluruhan, REDLAMP dapat diterapkan untuk mendeteksi anomali pada data PM2.5 univariat secara unsupervised, dengan performa optimal yang bergantung pada konfigurasi ukuran window dan jumlah kelas pseudo-anomali yang digunakan.
================================================================================================================================
In PM2.5 data from air quality monitoring stations (SPKU), time-series anomaly detection is needed to flag points or subsequences that deviate from normal patterns before the data is used in further analysis, and because the data used does not necessarily provide anomaly labels, an unsupervised approach that can detect deviations based on patterns learned from the data is required. Existing unsupervised approaches still have limitations in handling anomaly contamination, diversity gap, false anomalies, and overconfidence in labels. To address these issues, this research applies REDLAMP (Robust and Explainable Detector of Time-Series Anomaly), an unsupervised method that combines reconstruction and classification with a multiclass approach, to detect anomalies in PM2.5 concentration time-series data from the Gelora Bung Karno air quality monitoring station under the Ministry of Environment. Through the combination of reconstruction error and adjusted anomaly-class score, the model produces an anomaly score that indicates the degree of deviation of a window from the learned pattern, and REDLAMP's performance is evaluated based on variations in window size and types of pseudo-anomalies resulting from augmentation. The evaluation results show that REDLAMP can distinguish anomalous points from normal points, with the highest VUS-ROC of 0.8489 and the highest VUS-PR of 0.1800. The reconstruction-only component consistently achieves the highest performance compared to the combined and classification-only components, longer windows reduce the model's ability to rank anomalies globally, and a configuration of six pseudo-anomaly classes (K=6) achieved the highest performance compared to K=2 and K=12. Overall, REDLAMP can be applied to detect anomalies in univariate PM2.5 data in an unsupervised manner, with optimal performance depending on the window size configuration and the number of pseudo-anomaly classes used.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deteksi Anomali, Time Series, PM2.5, REDLAMP, Pseudo-Anomali Multi Kelas, Anomaly detection, time-series, REDLAMP, multiclass pseudo-anomaly, PM2.5 concentration.
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Alendra Rafif Athaillah
Date Deposited:	26 Jul 2026 23:32
Last Modified:	26 Jul 2026 23:32
URI:	http://repository.its.ac.id/id/eprint/137432

Actions (login required)

View Item