IPB University Logo

SCIENTIFIC REPOSITORY

IPB University Scientific Repository collects, disseminates, and provides persistent and reliable access to the research and scholarship of faculty, staff, and students at IPB University

AI Repository
 
Building and Categories


      View Item 
      •   IPB Repository
      • Final Assignments
      • Master Final Assignments
      • MF - School of Data Science, Mathematic and Informatics
      • View Item
      •   IPB Repository
      • Final Assignments
      • Master Final Assignments
      • MF - School of Data Science, Mathematic and Informatics
      • View Item
      JavaScript is disabled for your browser. Some features of this site may not work without it.

      Perbandingan Kinerja ST-DBSCAN dan ST-SCKM Serta Model Hybrid dalam Pengelompokan Data Kualitas Udara Berbasis Spasial-Temporal

      Thumbnail
      View/Open
      Cover (604.9Kb)
      Fulltext (20.69Mb)
      Lampiran (509.9Kb)
      Date
      2026
      Author
      Idris, Muh. Akbar
      Aidi, Muhammad Nur
      Djuraidah, Anik
      Metadata
      Show full item record
      Abstract
      Penggerombolan data spasial-temporal diperlukan untuk mengenali pola yang berubah menurut atribut, lokasi, dan waktu. Metode berbasis kepadatan dapat membentuk gerombol tidak beraturan sekaligus menghasilkan label noise, tetapi sensitif terhadap perbedaan kepadatan dan parameter. Metode berkendala spasial memberikan regionalisasi yang lebih menyeluruh, tetapi tidak secara langsung mempertahankan kandidat anomali sebagai keluaran akhir. Penelitian ini membandingkan Spatio-Temporal Density-Based Spatial Clustering of Applications with Noise (ST-DBSCAN), Spatio-Temporal Spatially Constrained K-Means (ST-SCKM), dan dua model hybrid. Kajian simulasi menggunakan rancangan faktorial penuh 2! yang mengombinasikan geometri sirkular dan koridor, jarak antarpusat 3 dan 7 satuan, serta kontaminasi anomali 0% dan 10%. Setiap skenario terdiri atas 1.200 observasi, empat gerombol acuan, dan sepuluh replikasi yang dianalisis secara terpisah pada ruang fitur empat dimensi yang memuat X, Y, waktu, dan F1 atau F2. Hasil simulasi menunjukkan bahwa ST-SCKM memperoleh rataan Composite Regionalization Score sebesar 0,7233 dan skor tertinggi pada tujuh dari delapan skenario. Skenario 3 menghasilkan skor yang sama antara ST-SCKM dan kedua model hybrid setelah pembulatan. Evaluasi pada skenario terkontaminasi menunjukkan bahwa ST-DBSCAN memperoleh rataan F1-score deteksi sebesar 0,7863, sedangkan Hybrid 2 memperoleh nilai sebesar 0,7206. Kajian empiris menggunakan data pemantauan PM2.5, NO2, dan NOx di Jepang tahun 2024 yang diperoleh melalui OpenAQ. Pengolahan dengan median tiga jam menghasilkan sampel akhir sebanyak 9.998 observasi yang dianalisis pada ruang fitur delapan dimensi, yaitu bujur, lintang, waktu, sinus bulan, kosinus bulan, PM2.5, NO2, dan NOx. Pendekatan ST-SCKM menghasilkan empat zona dengan cakupan 100% dan Composite Zoning Score tertinggi sebesar 0,7919, sedangkan metode ST-DBSCAN menghasilkan cakupan 81,69% dengan 1.831 observasi berlabel noise. Hasil penelitian menempatkan metode ST-SCKM sebagai metode utama untuk regionalisasi dan metode ST-DBSCAN sebagai metode pendukung untuk mendeteksi kandidat anomali pada konfigurasi yang diuji. Empat zona hasil metode ST-SCKM memperlihatkan perbedaan relatif dalam profil polutan, lokasi, dan musim dominan, sementara pola High-High dan Low-Low dari analisis spasial lokal digunakan sebagai informasi deskriptif. Nilai F1-score metode ST-DBSCAN terhadap pseudo-acuan rentang antarkuartil mencapai 0,4443 untuk PM2.5, 0,4513 untuk NO2, dan 0,5313 untuk NOx, tetapi pemeriksaan sensor, meteorologi, sumber emisi, dan kejadian lokal tetap diperlukan untuk memastikan makna substantif setiap kandidat.
       
      Spatio-temporal clustering is required to identify patterns that vary across attributes, locations, and periods. Density-based methods can identify irregularly shaped clusters and retain noise labels, although their results are sensitive to density variation and parameter selection. Spatially constrained methods provide more complete regionalization but do not directly retain anomaly candidates as their final output result. This study compared Spatio-Temporal Density-Based Spatial Clustering of Applications with Noise (ST-DBSCAN), Spatio-Temporal Spatially Constrained K-Means (ST-SCKM), and two hybrid models. The simulation study employed a full 2! factorial design combining circular and corridor geometries, center separations of 3 and 7 units, and anomaly contamination levels of 0% and 10%. Each scenario contained 1,200 observations, four reference clusters, and ten replications analyzed separately in two four-dimensional feature spaces consisting of X, Y, time, and either F1 or F2. The simulation results showed that ST-SCKM method achieved a mean Composite Regionalization Score of 0.7233 and the highest score in seven of the eight scenarios. Scenario 3 produced equal rounded scores for ST-SCKM method and both hybrid models. Detection evaluation in the contaminated scenarios showed that ST-DBSCAN method achieved a mean F1-score of 0.7863, compared with 0.7206 for Hybrid 2. The empirical study used 2024 monitoring data for PM2.5, NO2, and NOx in Japan obtained through OpenAQ. Three-hour median aggregation produced a final sample of 9,998 observations analyzed in an eight-dimensional feature space consisting of longitude, latitude, time, month sine, month cosine, PM2.5, NO2, and NOx. ST-SCKM generated four zones with 100% coverage and the highest Composite Zoning Score of 0.7919, whereas ST-DBSCAN achieved 81.69% coverage and labelled 1,831 observations as noise. The findings support ST-SCKM as the primary regionalization method and ST-DBSCAN as a supporting method for candidate anomaly detection under the evaluated configurations. The four ST-SCKM zones exhibited relative differences in pollutant profiles, locations, and dominant seasons, while local High-High and Low-Low spatial patterns were interpreted descriptively. ST-DBSCAN achieved F1-scores of 0.4443 for PM2.5, 0.4513 for NO2, and 0.5313 for NOx against the interquartile-range pseudo-references, although sensor records, meteorological conditions, emission sources, and local events remain necessary to verify the substantive meaning of each candidate.
       
      URI
      http://repository.ipb.ac.id/handle/123456789/179455
      Collections
      • MF - School of Data Science, Mathematic and Informatics [179]

      Copyright © 2020 Library of IPB University
      All rights reserved
      Contact Us | Send Feedback
      Indonesia DSpace Group 
      IPB University Scientific Repository
      UIN Syarif Hidayatullah Institutional Repository
      Universitas Jember Digital Repository
        

       

      Browse

      All of IPB RepositoryCollectionsBy Issue DateAuthorsTitlesSubjectsThis CollectionBy Issue DateAuthorsTitlesSubjects

      My Account

      Login

      Application

      google store

      Copyright © 2020 Library of IPB University
      All rights reserved
      Contact Us | Send Feedback
      Indonesia DSpace Group 
      IPB University Scientific Repository
      UIN Syarif Hidayatullah Institutional Repository
      Universitas Jember Digital Repository