A Study on the Mid- to Long-Term Demand Forecasting for Beauty Arts Major Students Using Advanced Machine Learning and Time Series Analysis
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
(Purpose) The Purpose of this study is to develop a medium- to long-term forecasting model for beauty arts major student demand using advanced machine learning and time series analysis techniques, and to provide empirical evidence for policy responses to the declining school-age population crisis in Korean higher education. (Design/methodology/approach) This study employed multiple forecasting models including ARIMA, Prophet, LSTM, XGBoost, SVM, and Random Forest using 2010-2024 time series data from University Information, KEDI, and Statistics Korea. Independent variables included demographic factors (18-year-old population, school-age population), economic factors (youth unemployment rate, GDP per capita), educational factors (college enrollment rate), and industry factors (cosmetics export value, K-beauty search volume index). Hybrid models combining ARIMA with machine learning techniques were constructed, and performance was evaluated using MAE, RMSE, MAPE, and R² metrics with cross-validation and Diebold-Mariano tests. (Findings) The STL-ARIMA-XGBoost hybrid model demonstrated superior performance with MAPE 2.81%, RMSE 245.6 persons, and R² 0.974, showing 31.5% RMSE improvement over single models. SHAP analysis revealed that 18-year-old population (0.342) had the greatest impact, followed by school-age population (0.228) and cosmetics export value with 2-year lag (0.156). Forecasting results indicate that total freshmen enrollment will decrease from 6,030 persons in 2024 to 4,780 persons in 2030 (20.7% decrease) and 3,850 persons in 2035 (36.1% decrease) under the medium scenario. (Research implications or Originality) This study provides methodological advancement by systematically comparing multiple algorithms and developing hybrid models capturing both linear and nonlinear patterns. The findings empirically demonstrate the application of demographic transition theory and ensemble learning theory to educational demand forecasting. This study contributes by quantitatively identifying the 2-year lag effect of K-beauty industry growth on major selection and establishing a data-driven policy framework for managing declining student demand in beauty arts education.
Abstract
(Purpose) The Purpose of this study is to develop a medium- to long-term forecasting model for beauty arts major student demand using advanced machine learning and time series analysis techniques, and to provide empirical evidence for policy responses to the declining school-age population crisis in Korean higher education. (Design/methodology/approach) This study employed multiple forecasting models including ARIMA, Prophet, LSTM, XGBoost, SVM, and Random Forest using 2010-2024 time series data from University Information, KEDI, and Statistics Korea. Independent variables included demographic factors (18-year-old population, school-age population), economic factors (youth unemployment rate, GDP per capita), educational factors (college enrollment rate), and industry factors (cosmetics export value, K-beauty search volume index). Hybrid models combining ARIMA with machine learning techniques were constructed, and performance was evaluated using MAE, RMSE, MAPE, and R² metrics with cross-validation and Diebold-Mariano tests. (Findings) The STL-ARIMA-XGBoost hybrid model demonstrated superior performance with MAPE 2.81%, RMSE 245.6 persons, and R² 0.974, showing 31.5% RMSE improvement over single models. SHAP analysis revealed that 18-year-old population (0.342) had the greatest impact, followed by school-age population (0.228) and cosmetics export value with 2-year lag (0.156). Forecasting results indicate that total freshmen enrollment will decrease from 6,030 persons in 2024 to 4,780 persons in 2030 (20.7% decrease) and 3,850 persons in 2035 (36.1% decrease) under the medium scenario. (Research implications or Originality) This study provides methodological advancement by systematically comparing multiple algorithms and developing hybrid models capturing both linear and nonlinear patterns. The findings empirically demonstrate the application of demographic transition theory and ensemble learning theory to educational demand forecasting. This study contributes by quantitatively identifying the 2-year lag effect of K-beauty industry growth on major selection and establishing a data-driven policy framework for managing declining student demand in beauty arts education.