View Item 
      •   IPB Repository
      • Dissertations and Theses
      • Master Theses
      • MT - School of Data Science, Mathematic and Informatics
      • View Item
      •   IPB Repository
      • Dissertations and Theses
      • Master Theses
      • MT - 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.

      Evaluasi Kinerja Model Spatial Clustering Regression dan Spatial Clustering Coefficient dalam Penanganan Efek Spasial

      Thumbnail
      View/Open
      Cover (4.745Mb)
      Fulltext (9.252Mb)
      Lampiran (2.515Mb)
      Date
      2025
      Author
      Syam, Ummul Auliyah
      Djuraidah, Anik
      Syafitri, Utami Dyah
      Metadata
      Show full item record
      Abstract
      Efek spasial merupakan karakteristik penting dalam analisis data spasial yang terdiri dari efek dependensi spasial dan heterogenitas spasial. Heterogenitas spasial menunjukkan adanya keragaman antar lokasi sedangkan dependensi spasial menunjukkan adanya keterkaitan antar lokasi pengamatan. Spatial Clustering Regression (SCR) dan Spatial Clustering Coefficient (SCC) merupakan dua pendekatan baru dalam penanganan efek spasial yang dirancang untuk menangkap pola koefisien regresi yang tidak homogen dan cenderung menggerombol di ruang geografis. SCR menggabungkan regresi spasial dengan metode pengelompokan berbasis K-means dan fungsi penalti likelihood yang dimotivasi dari model Potts, sehingga memungkinkan identifikasi wilayah dengan karakteristik regresi serupa. SCC, di sisi lain, mengadopsi pendekatan penalti least absolute shrinkage and selection operator yang dikombinasikan dengan struktur pohon rentang minimum (minimum spanning tree) untuk mendorong keseragaman koefisien regresi antar lokasi yang berdekatan. Keduanya ditujukan untuk mengatasi keterbatasan metode spasial konvensional seperti Geographically Weighted Regression (GWR) yang cenderung menghasilkan penduga ekstrem dan sulit diinterpretasikan. GWR menghitung dugaan parameter regresi di setiap titik lokasi secara lokal, hanya berdasarkan pengaruh titik-titik terdekat dengan bobot spasialnya. Namun, GWR tidak secara eksplisit mengidentifikasi atau mempertimbangkan kelompok (gerombol) dalam data spasial. Penelitian ini mengkaji kinerja SCR dan SCC melalui kajian simulasi serta penerapan pada kasus stunting di Indonesia. Melalui kajian simulasi, dibangkitkan data spasial pada 1.000 lokasi dengan kombinasi tiga tingkat autokorelasi spasial (lemah, sedang, kuat) dan korelasi antar peubah penjelas (lemah, sedang, kuat), diulang sebanyak 100 kali. Evaluasi kinerja dilakukan dengan membandingkan rata-rata Root Mean Square Error (RMSE) hasil dugaan parameter antara SCR, SCC, dan GWR. Hasil simulasi menunjukkan bahwa SCR secara konsisten menghasilkan dugaan parameter yang paling mendekati nilai sebenarnya dan mampu mengidentifikasi pola gerombol spasial yang tajam dan realistis, terutama pada kondisi autokorelasi spasial sedang hingga kuat. SCC juga mampu menangkap pola gerombol namun cenderung overfitting, sedangkan GWR memberikan hasil terlalu halus dan kurang peka terhadap perubahan antarwilayah. Secara statistik, SCR memiliki RMSE lebih rendah di hampir semua skenario dibandingkan dua metode lainnya, dan perbedaan antar model signifikan berdasarkan uji Kruskal-Wallis. Kajian pada data empiris diterapkan pada data kasus stunting tahun 2022 pada 510 kabupaten/kota di Indonesia digunakan, dengan tujuh peubah penjelas yang meliputi tingkat kemiskinan, jumlah puskesmas, gizi ibu hamil, akses sanitasi, akses air bersih, produk domestik regional bruto (PDRB), dan rata-rata lama sekolah. Analisis awal menunjukkan adanya efek heterogenitas dan dependensi spasial, sehingga regresi spasial merupakan pendekatan yang tepat. SCR diterapkan dengan membandingkan lima jenis matriks pembobot spasial berbasis jarak, dan model terbaik diperoleh pada pembobot 3-tetangga terdekat dengan empat gerombol optimal. Hasil penggerombolan menunjukkan bahwa setiap wilayah memiliki karakteristik lokal berbeda yang mempengaruhi persentase stunting. Gerombol 1 meliputi wilayah dengan rata-rata persentase stunting terendah (20.31%), mencakup Pulau Jawa Bali, Kalimantan (Provinsi Kalimantan Barat, Kalimantan Tengah, Kalimantan Selatan), Provinsi Sumatera Selatan, Lampung, dan Kepulauan Bangka Belitung. Wilayah ini ditandai dengan tingkat kemiskinan paling rendah (8.61%), akses air bersih dan sanitasi yang tinggi (88.91% dan 81.50%), serta rata-rata lama sekolah yang cukup baik (8.33 tahun). Faktor dominan penurunan stunting di wilayah ini adalah pendidikan dan infrastruktur dasar yang memadai. Gerombol 2 mencakup sebagian besar Pulau Sumatera, dengan rata-rata stunting 24.49% dan PDRB tertinggi di antara semua gerombol. Meskipun secara ekonomi relatif baik, kemiskinan tetap menjadi faktor signifikan yang meningkatkan stunting. Rata-rata lama sekolah juga tinggi (9.08 tahun), namun terbatasnya jumlah puskesmas menjadi perhatian penting untuk intervensi kesehatan. Gerombol 3 mencakup wilayah Pulau Sulawesi, Provinsi Kalimantan Utara, Kalimantan Timur, Nusa Tenggara Barat, Nusa Tenggara Timur, dan Maluku Utara, dengan rata-rata persentase stunting sebesar 28.38%. Tingkat kemiskinan lebih tinggi dari gerombol sebelumnya (12.22%) dan memiliki akses air bersih dan pendidikan cukup baik, meskipun jumlah fasilitas kesehatan masih terbatas. Jumlah puskesmas dan tingkat kemiskinan menjadi dua faktor utama yang mempengaruhi stunting di wilayah ini. Gerombol 4 merupakan wilayah dengan persentase stunting tertinggi (33.16%), terdiri dari Provinsi Papua, Papua Barat, dan Maluku. Daerah ini memiliki karakteristik sosial ekonomi paling tertinggal, ditunjukkan oleh tingkat kemiskinan tertinggi (26.66%), akses sanitasi dan air bersih terendah (52,68% dan 73.90%), serta rata-rata lama sekolah hanya 7.10 tahun. Faktor dominan yang mempengaruhi stunting di wilayah ini adalah keterbatasan infrastruktur sanitasi dan pendidikan. Penelitian ini menunjukkan bahwa SCR tidak hanya efektif dalam mendeteksi pola spasial gerombol, tetapi juga mampu mengelompokkan wilayah berdasarkan faktor lokal yang relevan dalam menyarankan kebijakan berbasis wilayah yang lebih kontekstual. SCR juga terbukti sebagai pendekatan yang efisien dan lebih akurat dalam menangani data spasial berskala besar. Identifikasi faktor dominan di tiap gerombol memungkinkan intervensi yang lebih efisien dan tepat sasaran. Selain itu, hasil ini menunjukkan bahwa kebijakan penanggulangan stunting tidak dapat digeneralisasi secara nasional, melainkan harus mempertimbangkan struktur spasial dan kondisi lokal masing-masing wilayah. Temuan ini memberikan implikasi penting dalam perumusan kebijakan penanggulangan stunting yang berbasis wilayah. Kebijakan intervensi perlu disesuaikan dengan karakteristik gerombol agar lebih tepat sasaran, efisien, dan berdampak nyata pada penurunan angka stunting di Indonesia.
       
      Spatial effects are an important characteristic in spatial data analysis, consisting of spatial dependency and heterogeneity effects. Spatial heterogeneity indicates the existence of diversity between locations, while spatial dependency indicates the existence of interrelationships between observation locations. Spatial Clustering Regression (SCR) and Spatial Clustering Coefficient (SCC) are two new approaches in handling spatial effects designed to capture patterns of regression coefficients that are not homogeneous and tend to cluster in geographic space. SCR combines spatial regression with a K-means-based clustering method and a penalty likelihood function motivated by the Potts model, allowing the identification of regions with similar regression characteristics. On the other hand, SCC adopts a penalty LASSO approach combined with a minimum spanning tree structure to promote uniformity of regression coefficients between neighboring locations. Both are intended to overcome the limitations of conventional spatial methods, such as Geographically Weighted Regression (GWR), which tends to produce extreme and difficult-to-interpret estimates. GWR calculates the estimated regression parameters at each location point locally, based only on the influence of nearby points with their spatial weights. However, GWR does not explicitly identify or consider clusters in spatial data. This study examines the performance of SCR and SCC through simulation studies and application to the case of stunting in Indonesia. Through simulation studies, spatial data were generated at 1,000 locations with three levels of spatial autocorrelation (weak, moderate, strong) and correlation between explanatory variables (weak, moderate, strong), repeated 100 times. Performance evaluation was conducted by comparing the Root Mean Square Error (RMSE) of parameter estimates between SCR, SCC, and GWR. Simulation results show that SCR consistently produces parameter estimates that are closest to the true values and is able to identify sharp and realistic spatial cluster patterns, especially under moderate to strong spatial autocorrelation conditions. SCC is also able to capture cluster patterns but tends to be overfitting, while GWR provides smooth results that are too smooth and is less sensitive to inter-regional changes. Statistically, SCR had a lower RMSE in almost all scenarios than the other two methods, and the difference between models was significant based on the Kruskal-Wallis test. A study of empirical data applied to 2022 stunting case data in 510 districts/municipalities in Indonesia was used, with seven explanatory variables including poverty level, number of public health centers, nutrition of pregnant women, access to sanitation, access to clean water, gross regional domestic product (GRDP), and average years of schooling. Preliminary analysis showed the effects of spatial heterogeneity and dependency, making spatial regression an appropriate approach. SCR was applied by comparing five types of distance-based spatial weight matrices, and the best model was obtained under the 3-nearest neighbor weight with four optimal clusters. The clustering results show that each region has different local characteristics that affect the percentage of stunting. Cluster 1 includes the regions with the lowest average percentage of stunting (20.31%), covering Java, Bali, Kalimantan (West Kalimantan, Central Kalimantan, South Kalimantan), South Sumatra, Lampung, and Bangka Belitung Islands. This region is characterized by the lowest poverty rate (8.61%), high access to clean water and sanitation (88.91% and 81.50%), and good average years of schooling (8.33 years). Education and adequate basic infrastructure are the dominant factors in reducing stunting in this region. Cluster 2 covers most of Sumatra Island, with an average stunting rate of 24.49% and the highest GRDP among all clusters. Despite relatively good economic conditions, poverty remains a significant factor that increases stunting. The average number of years of schooling is also high (9.08 years), but the limited number of public health centers is an important concern for health interventions. Cluster 3 covers Sulawesi Island, North Kalimantan, East Kalimantan, West Nusa Tenggara, East Nusa Tenggara, and North Maluku, with an average stunting percentage of 28.38%. The poverty rate is higher than the previous cluster (12.22%), and it has good access to clean water and education, although the number of public health centers is still limited. The number of public health centers and the poverty rate are the two main factors affecting stunting in this area. Cluster 4 has the highest percentage of stunting (33.16%), consisting of Papua, West Papua, and Maluku Provinces. This region has the most underdeveloped socioeconomic characteristics, indicated by the highest poverty rate (26.66%), the lowest access to sanitation and clean water (52.68% and 73.90%), and an average years of schooling of only 7.10 years. The dominant factors influencing stunting in this region are limited sanitation infrastructure and education. This research shows that SCR effectively detects clustered spatial patterns and can cluster regions based on relevant local factors to suggest more contextualized region-based policies. SCR was also an efficient and more accurate approach to handling large-scale spatial data. Identifying dominant factors in each cluster allows for more efficient and targeted interventions. In addition, these results show that stunting prevention policies cannot be generalized nationally, but each region's spatial structure and local conditions must be considered. This finding has important implications for the formulation of area-based stunting prevention policies. Intervention policies need to be adjusted to the characteristics of the clusters to be more targeted and efficient, and have a tangible impact on reducing stunting rates in Indonesia.
       
      URI
      http://repository.ipb.ac.id/handle/123456789/161820
      Collections
      • MT - School of Data Science, Mathematic and Informatics [69]

      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