特徵價格法為常用判斷不動產價值及估價的四種方法之一，其被廣泛使用於具有影響不動產價格之特徵的都市地區。但不動產在判斷價值時，由於估價依個人主觀意識不同結果也不同，缺乏一套客觀的標準存在，因此本研究目的在於以特徵價格法為基礎，找出影響不動產環境特徵，並用支撐向量機法(SVM)建立不動產估價模型，進而消除由人為主觀判斷不動產價值的缺陷。本研究透過文獻蒐集影響住宅價格之特徵因子，並經整理篩選後共19個環境特徵因子，接著以台北市12行政區蒐集2007年到2010年之歷史交易資料。在隨機選取5000筆資料中，去除遺漏及空缺資料剩下4165筆。經統計分析和極端值剔除除後，刪除其中174筆資料。而測試組及訓練組使用隨機分類五組之交叉驗證(cross validation)概念。而在預測結果方面，以全台北市12行政區之交易資料進行分析，預測準確率為72.2%；而在案例分析部分，將南港區及內湖區交易資料放入預測模型中，其準確率為81.8%，表示本預測模具解釋住宅價格之能力。即SVM應用於估價模型是可行的方式，特別是對那些含有同性質特徵，且均勻分佈該區域內之地區。The hedonic pricing approach is one of four common methods used to price or predict real estate value. It is widely used especially for those in urban areas, which contain numerous features influencing price of real estate. Nevertheless this estimation is vulnerable due to human bias. The research objective is to establish an support vector machine (SVM) model, based on the hedonic pricing concept, to price urban real estate and, as such, to eliminate possible human bias made by appraisers. Literature review summarizes 19 features that commonly show up for the hedonic pricing approach. The data collection targets at historical housing transactions in Taipei city from 2008 to 2010. A total of 5000 randomly selected transactions are collected where only 4165 datasets are qualified due to omissive values in the data. Statistical analyses that examine correlation and abnormal values trim off other 174 datasets out of 4165. Using the 5-way cross validation concept for training and testing, the SVM model is developed to predict housing price. The results demonstrate that the proposed model reaches accuracy at 72.2% in the range of ±20% for the price. For two specific areas among all 12 districts in Taipei city, the model accuracy increases up to 81.8%. SVM pricing model is confirmed to be feasible especially for those districts containing features that are coessential and evenly spread out in the area.
