以類神經網路模式探討住宅房價影響因子之研究

Abstract

台灣房屋價格多數倚賴專家經驗判斷，但是依據條件每個人主觀的認知不盡相同，也由於交易資訊的不透明化，造成房價的不健全，所以要如何找到影響房價的因子，是許多研究的方向。本研究以倒傳遞類神經網路進行台北市南港區住宅房屋價格之預測。經文獻回顧方式，找出影響住宅房價的15個影響因子作為輸入變數。在輸入倒傳遞類神經網路前，先進行數據篩選及正規化，經過164組訓練案例及39組測試案例後，達到81.81％的精度，顯示房價是與影響因子存在某種關聯。最後將訓練結果再進行敏感度分析，找出實際影響台北南港區房屋價格之因子。研究結果顯示，對南港區房價較具有影響性的因子有3個，分別為總樓層數、建築物公設比、距離國中的遠近。然而在15項因子中敏感度分析相對最低的2個因子，刪除國小距離及景氣指標二個因子後，輸入類神經網路後完成收斂，顯示剩下的13個房價影響因子，對於類神經網路都有重要的影響性。

Taiwan housing prices in most rely expertise judgment, but on the basis of subjective cognitive conditions of each person are not the same. But also because the transaction information is not transparent, resulting in prices is not perfect. So how to find the factors that affect housing prices, many research direction. In this study, back-propagation neural network to predict the price of residential houses Nangang District, Taipei City. By way of literature review to identify the impact of residential housing prices 15 factor as input variables. Before entering back-propagation neural network, the first for data filtering and normalization, after 164 cases and 39 sets of training after a set of test cases, reached 81.81% accuracy, showing the price is associated with the existence of a factor. Finally, the training results then sensitivity analysis to identify the real impact factor Taipei Nangang District house price. The results show, for the Nangang District has an impact on property prices rose by more factors have three, respectively, where the total number of floors, the ratio of public buildings, and the distance between the two countries. However, in the 15 factors in the sensitivity of the two relatively low number of factors, the deletion of the small distance and the climate index two factors, the input type of neural network to complete the convergence, showing the remaining 13 house price impact factors for the class of neural networks The road has an important influence.