NEURAL NETWORKS FOR THE MASS APPRAISAL OF REAL-ESTATE

In this paper, a method for mass appraisal is presented which is based entirely on the estimation of value differences between properties. T he value difference between two properties is calculated by multiplica tion of the differences in the characteristics considered, with margin al adjustment factors for those characteristics. The marginal adjustme nts used in this method are calculated using artificial neural network s. Artificial neural networks are special computer programs which esta blish a network of very simple operating units. Each unit transforms i ts input according to the same function, resulting in an output The in put can come from outside the network (input unit) or from other units within the network. The amount to which the output of one unit contri butes to the input of another unit is controlled by the weights. The d efinition of these weights is done by ''training '' the network. The a rtificial neural network for appraisal can be trained with ''patterns '' of characteristics of properties and the market value of those prop erties. The training of the network can be considered as the market an alysis with the available sales data. Market analysis is an essential part of each mass-appraisal process. The market analysis using neural networks provides substantially more data to the appraisal process tha n other methods of market analysis.