Improving Regression Model Menggunakan Bagging MARS Terhadap Gini Ratio di Pulau Jawa

Abstract. Economic inequality in Java Island is a crucial issue that impacts sustainable development. This study aims to model the Gini ratio using the Multivariate Adaptive Regression Splines (MARS) method combined with Bootstrap Aggregating (Bagging). Secondary data includes the response variable Gini ratio and predictor variables such as GRDP, Open Unemployment Rate, Percentage of Poor Population, Population Growth, and Labor Force Participation Rate. The determination of the number of bootstrap replications (B = 10, 25, 50, 100, and 200), as well as model parameter settings for the maximum number of basis functions, maximum interaction, and minimum observation, were conducted. The best model was selected based on the replication with the lowest average Generalized Cross Validation (GCV) value and the highest R-squared value. The analysis results showed that the 100th replication produced the lowest average GCV, while the final model was derived from the 41st replication, which had the highest R-squared value. The final model consists of 17 basis functions, with the most influential variable being the percentage of the poor population. Based on the findings, it is crucial for the government to focus on reducing the percentage of the poor population through strategic policies to mitigate economic inequality in Java Island. Abstrak. Ketimpangan ekonomi di Pulau Jawa merupakan isu penting yang berdampak pada pembangunan berkelanjutan. Penelitian ini bertujuan untuk memodelkan gini ratio menggunakan metode Multivariate Adaptive Regression Splines (MARS) yang dikombinasikan dengan Bootstrap Aggregating (Bagging). Data sekunder yang digunakan mencakup variabel respon gini ratio serta variabel prediktor seperti PDRB, Tingkat Pengangguran Terbuka, Persentase Penduduk Miskin, Pertumbuhan Penduduk, dan Tingkat Partisipasi Angkatan Kerja. Penentuan jumlah replikasi bootstrap (B = 10, 25, 50, 100, dan 200), serta pengaturan parameter model juga dilakukan untuk jumlah maksimal basis fungsi, maksimum interaksi, dan minimum observasi. Model terbaik dipilih berdasarkan replikasi dengan nilai rata-rata Generalized Cross Validation (GCV) terendah dan nilai R-squared tertinggi. Hasil analisis menunjukkan bahwa replikasi ke-100 menghasilkan rata-rata GCV paling kecil, sementara model akhir yang diperoleh adalah dari replikasi ke-41 yang memiliki nilai R-squared tertinggi. Model akhir terdiri dari 17 basis fungsi, dengan variabel paling berpengaruh adalah persentase penduduk miskin. Berdasarkan hasil penelitian, penting bagi pemerintah untuk fokus pada pengurangan persentase penduduk miskin melalui kebijakan strategis guna mengurangi ketimpangan ekonomi di Pulau Jawa.