Preliminary Studies On Real Estate Appraisal Based On ANN

With the establishment and gradual perfection of the system of the real estate market in China, exercise of appraisal, as the agent in the real estate industry, plays an increasingly important role and serves in the increasingly broader scope in the real estate market. Real estate appraisal goes a long way to promote the normalization of the real estate price, to ensure the fair transaction in the real estate market, and to set up sound market system and reasonable order of transaction. Currently, the theories of real estate appraisal are still at the probing stage in our country; moreover, real estate appraisal calls for multi-dimensional development, therefore, the probe into the more scientific approaches is of both theoretical and practical significance.The thesis, on the basis of analyses and studies on the real estate appraisal in terms of its status quo at home and abroad, and the price property, and the factors influencing the real estate prices, systematically explicates three appraisal approaches widely-used at present — market comparison approach, income approach and cost approach, analyzing respectively the theoretical grounds, the application scope and the evaluation procedures of each approach. On the basis of the review and tests of plenty of relevant literature, the thesis introduces the theories of ANN into the field of real estate appraisal, and by employing its super-powerful abilities of learning and handling non-linear relationship, finds out the objective law (i.e. the law between the real estate price and its influencing factors) through the analyses of plenty of training samples — real estate de facto transactions, and establishes a model of ANN real estate appraisal.In the realization of the model of ANN real estate appraisal, this thesis, using MATLAB as a tool, sets up a three-layer neural network model with the factors influencing the real estate price as the input artificial neuron and the real estate prices as the output artificial neuron, in which the BP neural network is optimized by using Levenberg-Marquardt method. Satisfactory results are achieved in the training and confirmation of ANN model of real estate appraisal with a group of real estate transactions in the current market.Both the theories and practice indicate that the adoption of BP neural network and Levenberg-Marquardt optimization method can help achieve promptly the accurate and satisfactory results in the real estate appraisal. It is a set of scientific and efficient approach to real estate appraisal and indicates a significant development in theory in this field.