SVR與OLS在住宅價格預測正確率的比較

A Comparison of the Accuracy Rates of SVR and OLS in Housing Price Prediction

在多數預測住宅價格之研究中，主要運用迴歸方法進行預測並探討住宅屬性對於住宅價格之影響。近期許多研究領域應用支撐向量機（support vector machines）作為分析方法，因其可運用於分類及迴歸預測，已逐漸成為相當熱門的研究方法之一。在各種不同方法之比較上，支撐向量機具有良好之分類及預測績效。本研究運用支撐向量機中的支撐向量迴歸（supportvector regression）建立住宅價格預測模型，並與普通最小平方法進行預測績效之比較。本研究蒐集台北市2008年至2010年住宅交易資料，扣除遺漏及極端值後，資料總數為5,261筆。實證結果顯示，支撐向量迴歸之預測正確率高於普通最小平方法且預測誤差小於普通最小平方法，表示支撐向量迴歸的預測績效較佳。

The main objective of this research was to predict the prices of residences in Taipei City. However, there are many factors that affect housing prices; hence, in this research, which is based on hedonic prices theory and the related literature, the attributes affecting housing prices are summarized for use as research variables. In numerous past studies that sought to predict housing prices, regression analysis was the primary method used to make the predictions and to investigate the influence of residence attributes on housing prices. Recently, however, the use of Support Vector Machines (SVMs) has been adopted for such analyses in a variety of research fields, including for the forecasting of housing prices. Moreover, the use of SVMs has gradually become a very popular research method because SVMs can be used in classification and regression prediction. Compared with a variety of different methods, SVMs have been shown to have better classification and forecasting performance. In this study, Support Vector Regression (SVR) was used to set up a housing price forecast model, and this model was compared with the ordinary least squares (OLS) model in terms of forecasting performance. Residence transactions data from 2008 to 2010 for Taipei City were collected, and, after removing the omitted and extreme values, the total number of data points was 5,261. According to the results of the empirical analysis, the forecasting correctness of SVR was higher than that of OLS, meaning that SVR achieved better forecasting performance.