The Study on the Investigation and Predictive model of New Hosing Price for the Readjustment Area-The Case Studies in the Hu Wei Liao and the Jheng Zin Liao, Taninan

• Main Authors: Lee Chia Chang, 李佳璋
• Other Authors: 紀雲曜
• Format: Others
• Language: zh-TW
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/24969625508374840589

碩士 === 長榮大學 === 土地管理與開發學系碩士班 === 92 === Recently, the trade of the readjustment areas is very hot because that where became main developed zones in the city. The demand of hosing market of the Hu Wei Liao and the Jheng Zin Liao readjustment areas, where are close to the Tainan science-based industrial park and the Tainan technology industrial park, is considerable due to traffic are convenient. For this reason, the trade of these two adjustment area is not only very hot but also the most intensive areas in Tainan lately. The purposes of this price were to investigate price of new housing and its hedonic property through questionnaire and developed a predictive model of the housing price by the method of back-propagation artificial neural network（BANN）, that is high accurate learning, fast recalling speed and it can use input information containing noise. But BANN has problems of local minima and slow convergence, so in the paper added the Cauchy machine switching condition to solve these two problems in order to establish the housing price model of Tainan readjustment area at the same time. The Semi-log form model is the best one among four traditional multiple model for two readjustment areas. However, the mean absolute percentage error（MAPE）of Hu Wei Liao is 10.73% and Jhen Gzin Liao is 9.21%. About BANN, the learning rate of Hu Wei Liao is 0.05, momentum term is 0.9, learning cycle is 500 times, the nodes of hidden layer 4 is the best, the MAPE 8.91% is the lowest .The convergence condition of the models of the Jhen Zin Liao is not obvious, because the area is developing presently, still many cases are building, many residenters who have signed contract with building company did not move into. The number of survey equal to 99 samples is not enough because we can not visit them, and this makes BANN learning effect bad. The outlier of the traditional multiple regression model is not same as BANN model. If we use general statistics method or liner regression to judge it is outlier or not, it’s not suitable totally to BANN model. Thus, in the artificial neural network model , it’s not suitable to use statistics method and liner regression to judge, so in this paper we deleted them further. Through sample random choose, we can know the chosing process of learning sample .If we want to get the best network learning effect, we should add the samples which are inside error range to learning sample combination.