Design and Implementation of Property Valuation Model Using Artificial Neural Network
No Thumbnail Available
Date
6/6/2020
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Land and buildings owned by a person or legal body are termed as real property. Property valuation is a process of determining a cash estimate of a property's value for given market conditions and the value of a property changes with market conditions, Consequently, a property's value is often updated to reflect changes in market conditions, including for example, recent real estate transactions. Property price changes over time result from specific and general effects of economic and social forces. Currently, there is no automated technique that can valuate a given property. Rather, property valuation is done manually. Naturally, this manual process of property valuation is prone to bias, inconsistencies, and corruption (bribery).
The goal of this research work is to design and implement a property valuation model capable of predicting a given residential property price at current time using artificial neural network based on the known and currently used approaches of property valuation and Ethiopian law on residential property. A novel identification technique is proposed to identify the significant attributes which have highest impact on property price. A total of 43 attribute have been used and 24 significant attributes are identified to model property valuation.
The model is designed by employing unsupervised learning approach using Artificial Neural Network based on sales comparison, replacement cost and income capitalization property valuation approaches. In order to strength the dataset and to fill the gap of missing values in the dataset geo-database is integrated in the model. The network is trained and its performance is tested with different accuracy and loss functions and the performance is extended using optimization metrics. For the purpose of training and testing the model, a total of 20113 data have been collected from FDRE documents authentication and registration agency, Abay bank, Dashen bank, Ayat real estate, Noah Real Estate Plc. The training data is randomly allocated into training (70%) and testing (30%). The model achieved an overall success rate of property price prediction using significant attributes is 97.47% and 89% using all attributes.
Description
Keywords
Property, Artificial Neural Network, Property Valuation, Sales Comparison, Replacement Cost, Income Capitalization, Geo-Database, Valuation Model, Nearest Neighbor, Attribute Selection