Peramalan Tingkat Pengangguran di Indonesia Menggunakan Support Vector Regression dan Long Short-Term Memory

Kebijakan Pemerintah yang memberlakukan physical distancing selama pandemi COVID-19, mulai PSBB hingga PPKM Level 3-4, menyebabkan mobilitas masyarakat terbatas. Terbatasnya mobilitas masyarakat akibat pandemi diperkirakan akan berdampak pada kenaikan Tingkat Pengangguran Terbuka (TPT) di Indonesia. Oleh karena itu, penelitian ini dilakukan untuk meramalkan TPT di Indonesia menggunakan 2 pendekatan, yaitu univariate time series dan gabungan dengan metode Support Vector Regression (SVR) dan Long Short-Term Memory (LSTM). Pendekatan univariate dilakukan menggunakan data TPT sedangkan pendekatan gabungan dilakukan menggunakan data TPT, data official statistics, dan data Google Trends. Didapatkan bahwa data Tingkat Pengangguran menunjukkan trend menurun sebelum terjadi pandemi COVID-19. Pada pemodelan ARIMA didapatkan bahwa asumsi residual berdistribusi normal tidak terpenuhi meskipun telah menggunakan outlier. Hal ini menunjukkan bahwa data yang digunakan merupakan data nonlinear. Dari hasil peramalan menggunakan pendekatan univariate dan gabungan dengan SVR dan LSTM, didapatkan bahwa metode terbaik yang dapat meramalkan data Tingkat Pengangguran adalah metode SVR pada pendekatan gabungan dengan nilai parameter γ=83 dan C=81 yang menghasilkan nilai RMSE dan MAPE pada data out sample berturut-turut sebesar 0,31426 dan 4,55657. ====================================================================================================== Government policies that impose physical distancing during the COVID-19 pandemic, ranging from PSBB to PPKM Level 3-4, cause limited community mobility. Limited mobility of people due to the pandemic is expected to have an impact on the increase in the Unemployment Rate in Indonesia. Therefore, this study was conducted to forecast the Unemployment Rate in Indonesia using 2 approaches, namely univariate time series and combined with Support Vector Regression (SVR) and Long Short-Term Memory (LSTM) methods. The univariate approach is done using Unemployment Rate data while the combined approach is done using Unemployment Rate data, official statistics data, and Google Trends data. It was obtained that the Unemployment Rate data showed a downward trend before the COVID-19 pandemic. In ARIMA modeling it is found that the residual assumption of normal distribution is not met despite having used an outlier. This indicates that the data used is nonlinear data. From the results of forecasting using univariate and combined approaches with SVR and LSTM, it was found that the best method that can forecast unemployment rate data is the SVR method on a combined approach with parameter values γ = 83 and C = 81 which produces RMSE and MAPE values in data out sample of 0,31426 and 4,55657, respectively