Nowcasting Jumlah Penduduk dengan Metode Support Vector Regression (SVR) dan Multi-Output Support Vector Regression (M-SVR)

Vinahari, Riyan Zulmaniar (2023) Nowcasting Jumlah Penduduk dengan Metode Support Vector Regression (SVR) dan Multi-Output Support Vector Regression (M-SVR). Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6003211036-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 September 2025. Download (4MB) | Request a copy

Abstract

Data dan informasi statistik beserta proyeksinya merupakan dasar penting untuk perencanaan dan evaluasi pembangunan nasional di masa mendatang khususnya jangka menengah maupun jangka panjang. Data tersebut tidak dapat dipenuhi melalui sistem registrasi sehingga data-data diperoleh dari sensus dan survei yang dilakukan Badan Pusat Statistik (BPS) untuk digunakan sebagai dasar proyeksi. Metode proyeksi penduduk yang umum digunakan di berbagai negara dan masih menjadi metode proyeksi penduduk Indonesia adalah Cohort Component Model (CCM). Akan tetapi, metode ini mempunyai beberapa kelemahan sehingga diperlukan metode baru yang lebih akurat salah satunya melalui metode nowcasting. Analisis time series tidak hanya dilakukan menggunakan statistika klasik, tetapi juga machine learning (ML). Metode ML mempunyai performa lebih unggul dibandingkan dengan metode statistika klasik. Algoritma ML yang umum digunakan untuk nowcasting adalah Support Vector Regression (SVR) dan Multi-output SVR (M-SVR). SVR juga mampu menghasilkan performa bagus dalam proyeksi di berbagai macam bidang. Akan tetapi SVR hanya mampu menangani single output. Sedangkan M-SVR mampu menangani permasalahan regresi multi-output seperti jumlah penduduk di beberapa provinsi di Pulau Jawa yang saling berkorelasi. Data yang digunakan bersumber dari BPS dengan variabel output jumlah penduduk di Provinsi DKI Jakarta, Jawa Barat, Jawa Tengah, dan Jawa Timur. Variabel input yang digunakan yaitu jumlah pelanggan listrik kelompok rumah tangga, jumlah angkatan kerja, jumlah rumah tangga, kepadatan penduduk (jiwa/km2), PDRB komponen PKRT ADHK (miliar Rp). Periode data yang digunakan merupakan data tahunan dari tahun 1985 sampai 2022. Pemilihan model terbaik dilakukan dengan membandingkan nilai RMSE dan MAPE out sample nowcasting model SVR dan M-SVR. Hasil penelitian menunjukkan bahwa berdasarkan nilai kebaikan model, nowcasting model SVR mempunyai performa yang lebih baik dibandingkan model M-SVR yang ditunjukkan dari nilai MAPE out sample nowcasting model SVR yang lebih kecil dibanding model M-SVR untuk seluruh variabel output. Selain itu, proyeksi penduduk hasil dari metode nowcasting memberikan performa yang lebih baik dibandingkan dengan metode CCM.
===================================================================================================================================
Statistical data and information, as well as their projections are an important basis for planning and evaluation of future national development, especially in the medium and long term development. These data cannot be provided through the population registration system. Hence, the data obtained from censuses and surveys conducted by Badan Pusat Statistik (BPS- Statistics Indonesia). The common method is Cohort Component Method (CCM). CCM is widely used in many countries including Indonesia, which still be utilized by BPS. Unfortunately, this method has several drawbacks, and therefore, more accurate method is required to outperform CCM. One of these methods is nowcasting. Time series analysis can now be conducted not only using classical statistical but also machine learning (ML). Machine learning is a new method in statistical forecasting that shows an excellent performance compared to classical statistical methods. Machine learning algorithms that are commonly used in sequential analysis are Support Vector Regression (SVR) and Multioutput SVR (MSVR) where both of these algorithms can perform nowcasting. SVR is also capable of producing excellent projections in a variety of disciplines. However, SVR is only able to handle single output. Meanwhile, M-SVR is capable of handling multi-output regression problems such as the number of populations in several provinces in Java Island which are correlated with each other. The data used was obtained from the BPS, and the output variable is the population of DKI Jakarta, West Java, Central Java, and East Java Provinces. The variables used as inputs are the number of electricity consumers for the household group, the number of workers, the number of households, the population density (people/km2), and the GRDP of the PKRT ADHK component (billion Rp). The data period used spans from 1985 to 2022 with annual data. Comparing the SVR and MSVR models' RMSE and MAPE values enables the selection of the optimal model. The results showed that based on the quality of the model, the SVR nowcasting model outperformed the M-SVR model, as shown by the SVR nowcasting model's out sample MAPE value, which was lower for all output variables than MAPE value from M-SVR. In addition, the efficacy of the population projections derived from the nowcasting method is outperform than CCM method.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	machine learning, multi-output support vector regression, nowcasting, penduduk, support vector regression; machine learning, multi-output support vector regression, nowcasting, population, support vector regression
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis H Social Sciences > HA Statistics > HA31.3 Regression. Correlation Q Science > Q Science (General) > Q325.5 Machine learning. Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) T Technology > T Technology (General) > T57.8 Nonlinear programming. Support vector machine. Wavelets. Hidden Markov models.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49101-(S2) Master Thesis
Depositing User:	Riyan Zulmaniar Vinahari
Date Deposited:	26 Sep 2023 02:15
Last Modified:	26 Sep 2023 02:15
URI:	http://repository.its.ac.id/id/eprint/104274

Actions (login required)

View Item