Model Geographically Weighted Regression dengan Fungsi Pembobot Adaptive dan Fixed Kernel pada Kasus Kematian Ibu di Jawa Timur

Model Geographically Weighted Regression (GWR) merupakan pengembangan dari model regresi linier berganda yang dapat menghasilkan penduga parameter model yang bersifat lokal untuk setiap titik atau lokasi di mana data diamati. Model GWR dapat digunakan apabila data memenuhi asumsi heterogenitas spasial yang diakibatkan oleh perbedaan kondisi data antara satu lokasi dengan lokasi lain. Penelitian ini bertujuan untuk menentukan model GWR terbaik dengan pembobot adaptive kernel dan fixed kernel pada kasus kematian ibu di Jawa Timur tahun 2018. Data yang digunakan pada penelitian ini adalah data kematian ibu sebagai variabel respon dan rumah tangga berperilaku hidup bersih sehat, kunjungan ibu hamil dengan K4, ibu hamil mendapat tablet Fe3, persalinan yang ditolong tenaga kesehatan, serta jumlah fasilitas kesehatan sebagai variabel prediktor. Berdasarkan kriteria pemilihan model terbaik yang dilihat dari nilai AIC terkecil dapat disimpulkan bahwa model GWR dengan fungsi pembobot adaptive bi- square kernel merupakan model terbaik untuk data kematian ibu. Faktor yang mempengaruhi kasus kematian ibu berdasarkan pengujian parameter secara parsial yaitu kunjungan ibu hamil dengan K4 dan jumlah fasilitas kesehatan.

Random Fourier Feature Based Deep Learning for Wireless Communications

This paper provides analytical results on fixed kernel width based RFF based DL (RFF-DL). The derived analysis and the presented case-studies indicate the RFF-DL's robustness to kernel-width initializations, and offers improved convergence in the low-data regime.

Random Fourier Feature Based Deep Learning for Wireless Communications

This paper provides analytical results on fixed kernel width based RFF based DL (RFF-DL). The derived analysis and the presented case-studies indicate the RFF-DL's robustness to kernel-width initializations, and offers improved convergence in the low-data regime.

Pemodelan Kriminal di Jawa Timur dengan Metode Geographically Weighted Regression (GWR)

Criminality constitutes all kinds of actions that are economically and psychologically harmful in violation of the law applicable in the state of Indonesia as well as social and religious norms, while the criminal data is the number of cases reported to the police institution. The higher the number of complainants the higher the number of criminals in the region. The greater the risk the community represents the more insecure a region is. This study aims to obtain the best model affecting crime or crime in East Java. The number of crimes in this study is limited to the number of theft cases (whether ordinary theft, theft by force, theft with theft, and the theft of motor vehicles). In this study, we use the Geographically Weighted Regression (GWR) model because this method is quite effective in estimating data that has spatial heterogeneity (uniformity in location / spatial). In essence, the model parameters in GWR can be calculated at the observation location with the dependent variable and one or more independent variables that have been measured at the sites where the location is known, where criminal acts in the research conducted in East Java involves the effects of spatial heterogeneity, with fixed kernel weighting function. The results showed that the variables affecting criminality in East Java Province are population density, economic growth, Gini Ratio, and poverty.

ANALISIS GEOGRAPHICALLY WEIGHTED REGRESSION (GWR) PADA KASUS PNEUMONIA BALITA DI PROVINSI JAWA BARAT

Pneumonia adalah salah satu penyakit Infeksi Saluran Pernapasan Akut (ISPA) yang disebabkan oleh bakteri atau virus. Di Indonesia, pneumonia merupakan penyebab kematian balita tertinggi kedua setelah diare. Kasus pneumonia pada balita terbanyak yang ditemukan di Indonesia adalah di provinsi Jawa Barat. Untuk mengatasi pneumonia balita perlu dianalisis faktor penyebab pneumonia balita. Salah satu cara untuk menganalisis faktor tersebut adalah dengan menggunakan analisis Geographically Weighted Regression (GWR). Analisis GWR merupakan pengembangan dari analisis regresi linier berganda yang dapat mengatasi keragaman wilayah/heterogenitas spasial sehingga menghasilkan model dan pendugaan parameter berbeda untuk setiap wilayah amatan. Hasil analisis GWR dengan menggunakan pembobot spasial \textit{Fixed Kernel Gaussian} menunjukkan bahwa model GWR lebih baik daripada model regresi linier berganda. Hal ini berdasarkan nilai R2, nilai R2 dari model GWR (88.34%) lebih besar dari R2 dari model regresi linier berganda (71.86%). Sementara, Jumlah Kuadrat Galat (JKG) untuk model GWR diperoleh 3.031 lebih kecil bila dibandingkan dengan nilai JKG dari regresi linier berganda yang sebesar 7.317. Secara umum, terdapat tiga faktor yang berpengaruh signifikan terhadap kasus pneumonia balita di provinsi Jawa Barat tahun 2014 yaitu jumlah balita gizi buruk, persentase bayi yang diimunisasi dasar lengkap dan jumlah puskesmas.

PEMODELAN DATA KEMATIAN BAYI DENGAN GEOGRAPHICALLY WEIGHTED NEGATIVE BINOMIAL REGRESSION

Overdispersion phenomenon and the influence of location or spatial aspect on data are handled using Binomial Geographically Weighted Regression (GWNBR). GWNBR is the best solution to form a regression analysis that is specific to each observation’s location. The analysis resulted in parameter value which different from one observation to another between location. The Weighting Matrix Selection is done before doing The GWNBR modeling. Different weighting will resulted in different model. Thus this study aims to investigate the best fit model using infant mortality data that is produced by some kind of weighting such as fixed kernel Gaussian, fixed kernel Bisquare, adaptive kernel Gaussian and adaptive kernal Bisquare in GWNBR modeling. This region study covers all the districts/municipalities in Java because the number of observations are more numerous and have more diverse characteristics. The study shows that out of four kernel functions, infant mortality data in Java2012, the best fit model is produced by fixed kernel Gaussian function. Besides that GWNBR with fixed kernel Gaussian also shows better result than the poisson regression and negative binomial regression for data modeling on infant mortality based on the value of AIC and Deviance. Keywords: GWNBR, infant mortality, fixed gaussian, fixed bisquare, adaptive gaussian, adaptive bisquare.

Adaptive radial sinc kernel distribution and its application in mechanical fault diagnosis

Based on the deficiency of fixed-kernel in the traditional time–frequency distribution, which is lack of adaptability, a new adaptive kernel function, which is named as the adaptive radial sinc kernel, is proposed according to design criteria of adaptive optimal kernel. The definition and algorithm of radial sinc kernel are given, and the proposed method is compared with the tradition time–frequency distribution. The simulation results show that the proposed method is superior to the traditional fixed-kernel functions, such as Wigner–Ville distribution, Choi–Williams distribution, cone-kernel distribution and continuous wavelet transform. The adaptive radial sinc kernel can overcome the deficiency of fixed-kernel function in traditional time–frequency distribution, adopt the optimizing method to filter the cross-terms adaptively according to the signal distribution, obtain good time–frequency resolution and has extensive adaptability for an arbitrary signal. Finally, the proposed method has been applied to the fault diagnosis of rolling bearing, and the experiment result shows that the proposed method is very effective.

Guiana dolphin home range analysis based on 11 years of photo-identification research in a tropical estuary

Home range studies provide significant insights on social organization and interactions, limiting resources and habitat use. Knowledge on home range and habitat use by Guiana dolphins, Sotalia guianensis , is still scarce. The aim of this study was to identify and analyze individual’s home ranges of Guiana dolphins in the Cananéia Estuary (~25°03′S, 47°55′W), located in southeastern Brazil. Photo-identification efforts were conducted between 2000 and 2010. From a total of 135,918 pictures taken, 34,086 (25%) were useful for individual identification. Two-hundred and five individuals were cataloged based on permanent notches along dorsal fin borders. Of the cataloged individuals, 31 had been identified a minimum of 20 times, on distinct dates, prior to this analysis. Home ranges were estimated for these individuals using 4 methods: minimum convex polygon (MCP), adaptive kernel with least-squares cross-validation (AKLSCV), fixed kernel with reference bandwidth (FKHREF), and fixed kernel with least-squares cross-validation (FKLSCV). The sizes of the estimated home ranges varied between 2.2 and 43.8 km 2 ( X¯ = 17.5 km 2 ) with MCP, between 0.8 and 82.5 km 2 ( X¯ = 15.6 km 2 ) with AKLSCV, between 3.9 and 244 km 2 ( X¯ = 72.4 km 2 ) with FKHREF, and from 0.6 to 70.6 km 2 ( X¯ = 13.5 km 2 ) with FKLSCV. Significant differences in size and shape of the generated areas were detected when comparing the 4 tested methods. Variation of individual’s home range sizes and an extensive overlap among home ranges of different Guiana dolphins in the Cananéia Estuary provide evidence that the region supports important resources for this species. Therefore, preventing habitat loss in this region is essential to guaranteeing the persistence of this population. O estudo sobre o uso de área pode fornecer informações sobre organização social e interações, recursos limitantes e sobre o uso de habitat. Ainda é escasso o conhecimento sobre como o boto-cinza, Sotalia guianensis usa seu habitat. O objetivo deste estudo foi identificar e analisar as áreas de uso individuais de S. guianensis no estuário de Cananéia (~25°03′S; 47°55′W), localizado no sudeste brasileiro. Os esforços de foto-identificação foram realizados entre os anos de 2000 e 2010. De um total de 135.918 fotografias tomadas, 34.086 (25%) foram úteis para identificações individuais. Duzentos e cinco indivíduos foram catalogados através das marcas permanentes presentes em suas nadadeiras dorsais. Dentre os indivíduos catalogados, 31 foram identificados ao menos em 20 ocasiões, em dias distintos, antes destas análises. As áreas de uso foram estimadas para estes indivíduos utilizando quatro métodos distintos: mínimo polígono convexo (MPC), kernel adaptativo com largura determinada pelo método de validação cruzada de quadrados mínimos (AKLSCV), kernel fixo com largura de referência (FKHREF) e kernel fixo com largura determinada pelo método de validação cruzada de quadrados mínimos (FKLSCV). Os tamanhos das áreas de uso geradas variaram entre 2,2 e 43,8 km 2 ( X¯ = 17,5 km 2 ) com o uso do MPC, 0,8 e 82,5 km 2 ( X¯ = 15,6 km 2 ) com o uso do AKLSCV, 3,9 e 244 km 2 ( X¯ = 72,4 km 2 ) com o uso do FKHREF, e de 0,6 a 70,6 km 2 ( X¯ = 13,5 km 2 ) com o uso do FKLSCV. Foram detectadas diferenças significativas no tamanho e formato das áreas geradas pelos quatro métodos testados. Variações nos tamanhos das áreas de uso individuais e a extensa sobreposição entre diferentes áreas de uso dos botos-cinza no estuário de Cananéia fornecem evidências de que a região provê recursos importantes para esta espécie. Portanto, prevenir a perda de habitat na região é essencial para garantir a persistência dessa população.