目標：本文主要是利用類神經網路的預測能力，進而達到預測醫療院所門診失約情況之目的。方法：本文主要以類神經網路、迴歸分析與星期特性移動平均法預測醫療院所門診失約之情況，並比較三種預測方法何者預測能力較佳。結果：本文不只採用倒傳遞類神經網路進行門診失約預測，也使用星期特性移動平均模式、迴歸分析等預測方法進行績效分析評估，分別以MSE與MAPE兩種方式的績效指標作比較，藉以評估何種方式之預測能力較佳。實證結果為倒傳遞網路在預測顧客失約人數的兩項績效指標皆優於星期特性移動平均、迴歸分析，因而得到倒傳遞網路較其兩者預測方法更準確之結果；其中三種預測方法的正確率比較：倒傳遞類神經網路約55.91%；星期特性移動平均約47.24%；迴歸分析約48.82%。另外當過度預測某一位病患時所喪失的機會成本大於預測某一位病患時會發生的資源成本0.71倍，倒傳遞網路的預測力優於迴歸分析；在機會成本小於資源成本7.25倍，倒傳遞網路的預測能力比星期特性移動平均佳；若機會成本大於資源成本3.16倍，迴歸分析的預測能力優於星期特性移動平均。結論：以預測值的誤差之MSE及MAPE來判斷，倒傳遞網路比其他兩者預測方法準確。

Objectives: This research used a measurement of Back-Propagation Networking (BPN) to forecast missed hospital appointments. Methods: We compared BPN with Day-of-week Moving Average and a Regression model. Results: BPN predicted mean squared error (MSE) and mean absolute percentage error (MAPE) better than the others. The forecasting accuracy of BPN was 55.91%, Day-of-week Moving Average 47.24%, and regression 48.82%. On cost analysis: (1) the opportunity cost between Regression and BPN is 0.71 times larger than the resource cost. (2) The opportunity cost between Day-of-week Moving Average and BPN is 7.25 times smaller than the resource cost. (3) The opportunity cost between Day-of-week Moving Average and Regression is 3.16 times larger than the resource cost. Conclusions: When MSE and MAPE are used to compare the performance of these forecasting methods, our results showed that BPN was better than Dayof-Week Moving Average and Regression.