In recent years, a smal number of research groups have investigated the applciation of neural network technology to real estate appraisal. the majority of these studies have concentrated on homogeneous areas (that is areas where properties are subject to the same environmetnal and locational factors). This is generally done to restrict the data set to one local sub-market. However, the models created are specialised and not locationally portable. This paper presents results indicating that features extracted from Census data can provide location surrogates that significantly improve prediction accuracy.
展开▼