Arquitetura híbrida inteligente para classificação de liquidez imobiliária urbana em leilões

Abstract

The role of liquidity in the real estate market has attracted attention in the financial literature because of its strong impact on the economy and the sectors it covers. The liquidity of a property is an indicator of the speed or the degree of ease with which properties are traded, traded and converted into monetary value. Much of the information on these properties is available in large internet databases. If on the one hand, access to real estate data is not a problem, extracting knowledge from these databases is. Knowledge Discovery in Data Bases (KDD) systems are applied as a solution for the extraction of knowledge in decision making in a risky condition in the real estate business, since it is uncertain to establish a limit for this negotiation. Decisions occupying a central space in organizations become more complex under conditions of uncertainty. This implies that to meet the demand for success and quality of decisions, a decision-making process must be established that will have as central elements, the scenarios of these decisions, the alternatives and their impacts. Therefore, the following general objective was defined: to evaluate intelligent techniques and develop an Intelligent Hybrid Architecture (AHI) for classifying urban real estate liquidity in auctions, supporting the decision making process with a Double Impact and Probability Matrix (MPID). To achieve this goal, a series of experiments were conducted applying intelligent techniques to an actual database on an Auction site, containing auctioned and non-armed properties, in the years 2016 to 2020, collected randomly. The evaluation of intelligent techniques for data mining such as Randon Forest (RF), Decision Tree and Multilayer Perceptron Neural Network (MLP), determined the most promising techniques and most adherent to the real estate data collected, in joint action with AHI. The main characteristic of AHI is its ability to predict discount values, exposure time, number of bids and the classification of the end product. Therefore, the proposed model is capable of predicting and classifying the liquidity of auction properties through the enrichment of the database, reducing the decision bias for the classification of real estate liquidity in auctions. The synergy between AHI and MPID made it possible to map the threats and also the opportunities in this sector. A new concept called bidding edge was created in this work, which determines the convergence of real bids for a property. The evaluation of the extracted knowledge is useful and can be applied in the auction and banking sectors. The solution developed reached 75% Score, training the AHI RF technique with the number of standard trees in the "randomForest" library with 500 trees.