房屋價格趨勢分析 使用類神經網路與自適性差異演化法

Abstract

不動產價格由於具有其位置的固定性、物件的特別性，因此較不易形成交易匯集的市場。不動產交易價格會依其個別條件，經由買賣雙方協議後產生交易，因此與一般商品依生產成本定價的狀況會有差異。不動產交易價格與不動產估價存在著資訊不平等的價格差異，資訊科技的日新月異，公開不動產交易資訊也成為了各國在開放政府資料的項目之一，交易資訊透明化讓不動產市場價格的趨勢更為顯著。 本研究透過實價登錄不動產交易資料將目標鎖定「住用」不動產，並且排除低報、高報等異常物件，使用房屋特徵項目進行 K-means 分群，取出各群在各時間單位下的「單價每平方公尺」的平均值，並依據此平均值進行時間序列分析。本實驗收集了103 年 8月到104 年 2月的實價登錄資料，使用R語言建構趨勢分析模型，使用訓練集通過自適性差異演化法的結果產生倒傳遞類神經網路的權重值，並以此作出各房屋群的漲跌預測，以利於觀察與呈現房屋買賣價格趨勢。

Real estate market has its own method for pricing depends on each property’s location、type and characteristics etc. It makes the difficulty for predict the trend of real estate market. In recent years, the Taiwan Government started to promote the Real Price Registration System to record all data of real estate transaction publicly. This RPR system is used to intensify the price transparency of real estate market. With higher transparency, the prediction model could be made more effective and reliable. In this research, we would like to build up a prediction model of house pricing which is made by RPR transaction data. We focus on the properties which are domestic estate in Taipei City and were traded from August 2014 to February 2015. Using the features, for example, address、type of building、floor made a K-Means clustering model to recognize different groups of buyers and sub-market. The prediction model is made for each sub-market by using modified neural network and time series methods.