整合類神經網路和田口方法建構不動產估價模式-以台中市房地產交易為例

Integrating Artificial Neural Network and Taguchi method on constructing the Real Estate Appraisal Model - An Example of Real Estate Market in Taichung City

鍾翔宇(Siang-Yu Jhong)

指導教授 ： 陳牧言

共同指導教授 ： 范敏玄(Min-Hsuan Fan)

國立臺中科技大學/資訊與流通學院/資訊管理系碩士班/碩士(2014年)

https://doi.org/10.6826/NUTC.2014.00122

摘要

由於房地產的交易價格會受到地點、坪數、樓層、交易時間等影響，而其中也包含著許多看不見的利益與其他干擾因素，因此要克服預測不動產交易行情是一件複雜且非線性的問題。最近十幾年來有不少學者試著以決策樹、支援向量機、類神經網路或線性迴歸等方法來探討房地產價格，從歷史資料中預測出房地產交易行情，但是房地產交易行情影響因素非常多，因此選用適當的條件因素非常重要，此研究探討傳統統計方式的主成份分析(PCA)和數理邏輯推論的Information Gain和Gain Ratio做為變數選取的方式。

本研究將以近年來相當流行的倒傳遞類神經網路（Back-propagation neural network），搭配田口方法（Taguchi method）中的直交表，由系統規劃出不同的參數組合，在不同水準的直交表中找出相對最佳的參數組合，讓類神經網路之準確率達到最佳狀態。而實驗結果也證明本研究所提出之方法較傳統的機器學習如如決策樹、線性迴歸等有較佳之預測成效。

最後透過實驗結果證明，傳統的統計方式在探討變數較多且相關性較高的資料集，較難找出其相關聯規則，本研究提出之模型具有最佳之預測成效，也成功大幅降低模擬運算的時間成本，在較少的實驗次數中更有效率找到最佳之預測成效，進而幫使用者預測出較接近目前房地產交易行情。

關鍵字

類神經網路 ； 田口方法 ； 不動產估價 ； 房地產市場

並列摘要

The real estate transaction prices will be affected by location, size, floor, trading time and so on. In addition, many invisible benefits and other confounding factors are also involved. Therefore, to overcome the problem of predicting real estate transaction prices is a complex and nonlinear issue. In the past decade, many scholars tried to explore the real estate prices by using decision trees, support vector machines, neural networks or linear regression methods and to predict the real estate transaction prices from historical data. Since the real estate transaction prices are affected by many factors, the choice of appropriate conditions is very important. Therefore, this study used principal component analysis (PCA) of traditional statistical methods and information gain and gain ratio in mathematical logic inference as variable selection approach.
This study used the back-propagation neural network, which is very popular in recent years, and the orthogonal table in the Taguchi method to come up with different combinations of parameters by the system. The best combination of parameters in different levels of orthogonal tables was recognized, so that the accuracy of the neural network would reach its utmost. The experimental results presented in this study also demonstrated that the method used by this study had better predictive results than traditional machine learning methods such as decision trees, and linear regression.
Finally, it also showed that the model proposed in this study had the optimal predictive effect, and could significantly reduce the cost of time in simulation operation. The best predictive results could be found with a fewer number of experiments more efficiently. Thus users could predict a real estate transaction price that is not far from the current actual prices.

並列關鍵字

Artificial Neural Network ； Taguchi method ； Real estate appraisal ； Real estate market

參考文獻

行政院全球資訊網. (2011). 特種貨物及勞務稅條例. 提供行政院官方發佈的新聞

蔡瑞煌, 高志明, & 張金鶚. (1999). 類神經網路應用於房地產估價之研究. 住宅學

Aizerman, A., Braverman, E. M., & Rozoner, L. I. (1964). Theoretical foundations of the

potential function method in pattern recognition learning. Automation and Remote

Control, 25, 821-837.