摘要
近年來，少子化已成為全世界的關鍵議題，台灣社會也面臨少子化的挑戰。依據國發會統計指出台灣最快在2022年可能出現「人口負成長」的狀況，將來可能導致勞動年齡人口的減少，平均每位工作人口負擔上、下兩代依賴人口的經濟壓力亦將更為沈重，社會安全無法獲得保障，未來的社會結構、經濟發展將會受到嚴重的影響。本研究提出結合粒子群最佳化與支持向量迴歸法應用於預測台灣每月人口出生數，研究資料來自內政部近20年出生人口數。研究結果顯示所提方法之平均絕對百分比誤差（Mean Absolute Percent Error, MAPE）低於自迴歸移動平均模型、指數平滑、向量自迴歸移動平均、向量自迴歸移動平均與外生迴歸變量與支持向量迴歸等方法。透過時間序列模型預測取得低誤差預測分析結果，以提供國家政府及相關單位參考，有助於解決人口負成長之國安危機。
並列摘要
In recent years, low fertility has become a key issue in the world, and Taiwan society has also been facing the challenge of declining birthrate. Taiwan's National Development Council pointed out that the nation's population might begin to decline in 2022, which may lead to less working-age population in the future; this results in a heavier economic pressure on the working population from both older and younger generations, social security not guaranteed, and social structure and economic development severely affected. This study proposes to combine both particle swarm optimization (PSO) and support vector machine (SVM) methods to predict the monthly fertility in Taiwan. With 20 years of data collected from Taiwan's Ministry of Interior, the analysis shows that the proposed method PSOSVR's mean absolute percent error is lower than the other methods such as ARIMA, ETS, VARMA, VARMAX, and SVR model. Minimal prediction error analysis via the time series forecasting model effectively provides a reference for the national government and relevant units to solve the national crisis of population growth.