本研究主要目的係針對臺北市中古屋大廈使用倒傳遞類神經網路（Back Propagation Neural Networks, BPNN）和調適性網路模糊推論系統（Adaptive Network-Based Fuzzy Inference System, ANFIS）進行價格預測，以提高房地產價格預測的品質。本研究所使用的變數可分為明確、模糊變數兩種。主要研究流程有三項：（1）針對臺北市不分群進行預測，（2）透過集群一、二的變數篩選找出最佳明確變數組合，（3）以內湖區、大安區為例，比較加入模糊變數(生活機能、周邊環境條件、預期發展潛力) 的與否之影響。主要的成果有三項：（1）僅以單一行政區進行預測的效果較臺北市不分群、分群來的良好，（2）內湖區加入模糊變數進行預測的MAPE（Mean Absolute Percentage Error）為6.15%，是本研究最佳的預測效果，（3）ANFIS 模型在內湖區以明確變數、加入模糊變數兩種組合進行預測的效果皆較BPNN來的良好，且更能有效地對模糊變數進行訓練與預測。

In pricing of real estates, subjective opinions can never be overlooked. As Fuzzy interpretation has demonstrated its strength in translating subjective opinions into quantified variables, it is now a handy tool when problems involve fuzziness. The major purpose of this paper is to introduce fuzzy variables to forecast the price of pre-owned house in Taipei City. For comparison, two methods, namely Back Propagation Neural Networks (BPNN) and Adaptive Neuro-Fuzzy Inference System (ANFIS), are implemented. After selection of variables, we include 5 Crisp and three Fuzzy variables in our model. Data from two Districts in Taipei city are analyzed and compared, we conclude that (1) Prediction with single administrative district is better than pooling all data; (2) Introducing Fuzzy variables improves the prediction significantly; (3) ANFIS outperformed BPNN in all aspects.