The Application of Intelligent Techniques for Massreal Estate Appraisal

Transcription

1 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, UDK 528.4: :332.74:510.6: Struèn èlanak The Applcaton of Intellgent Technques for Massreal Estate Apprasal Mroslav KUBURIÆ Subotca 1, Goran ÆIROVIÆ Belgrade 2 ABSTRACT. The paper revews the concept of mass apprasal of real estate, wthn whch, besdes defnng the basc concepts, a comparatve analyss s carred out of dfferent nternatonal experences related to ths ssue. The normatve and nsttutonal order of the subect area s analyzed n a test area, whle the concept of evaluatng spatal unts as a base of massreal estate apprasal and the feld of ther use are also defned. The essence of a valuaton model of spatal unts s defned, based on the prncple of case based reasonng (CBR) and logcal aggregaton (LA), and the mathematcal bass for the proposed model s gven for antcpatng the average prce of real estate wthn spatal unts. In the proposed model, spatal unts are descrbed, that s, the method of ther normalzaton and ther granulaton nto groups. Indvdual attrbutes and groups are allocated approprate weght by whch ther ndvdual and group sgnfcance are defned wthn the framework of the ntegral model, and fnally, testng of the model was carred out n a test area. Keywords: evaluaton of spatal unts, mass apprasal of real estate value, case-based reasonng, logcal aggregaton. 1. Introducton One of the fundamental reasons for establshng cadastral records, and thus the generator of geodesy, s the obectve and far admnstraton of tax polcy. From ts concepton to the present, the natonal spatal data nfrastructure s one of the most mportant factors n the mplementaton of ths polcy. Adustng to the real needs of ther users, wthout devatng from ther basc prncples, cadastral records, partcularly n developed countres, are constantly evolvng. Valung real estate s one of ts attrbutes whch has been expandng more and more n recent years. 1 Mroslav Kuburæ, PhD, Geod. Eng., Faculty of Cvl Engneerng Subotca, Unversty of Nov Sad, Kozaraèka 2a, RS Subotca, Serba, e-mal: mkuburc@gf.uns.ac.rs, 2 Prof. Goran Ærovæ, PhD, Cv. Eng., Faculty of Techncal Scence, Unversty of Nov Sad, Trg Dostea Obradovæa 6, RS Nov Sad, Serba, e-mal: crovc@sezampro.rs.

2 40 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, Real estate n the development of human cvlzaton also has partcular sgnfcance and can be consdered as mult-spectral. Ownershp of land or buldngs, or ther use, has a drect exstental connecton wth ther owner. That connecton s often materalzed as a relatonshp wth the hstorcal past or famly nhertance, exstental securty, or poston. Alone ts value s, n broad terms, the monetary value whch the real estate can make on the market at the moment of ts sale, and on the other hand t s the total resultng value of all of the most sgnfcant attrbutes that affect ts value. Astronomcal leaps and unexpected collapse of the real estate market are ust a few of the factors whch support the thess that real estate value s an ntegral functon of the value of ts attrbutes, some of whch often reman unknown even to the greatest experts. One of the precondtons to successfully modellng the process of mass apprasal of real estate value s defnng the feld of ts applcaton, and selectng the crtera and methodology for the apprasal are drectly related to the sze of ts doman. Namely, the greater the populaton of buldngs for the mass apprasal, as a result of the expanson of ts feld of applcaton, the more complcated and complex the number of crtera, the methodology and the modellng become. Hence, the ntegral model must be broken down nto ts own subsystems, n order for t on one hand to be unversally applcable, and on the other hand, to a model whch not only recognzes the relevant attrbutes of each property beng valued, but also assgns characterstcs to the area n whch that property s located. The unversal model of mass apprasal of real estate value n a partcular country must n essence be functonal, practcally applcable, consstent and adaptable to the real condtons and trends n the real estate market. It must also recognze all relevant factors whch nfluence the prce of real estate n each spatal unt, and at the same tme preserve all of the essental features of that area and use them n the process of determnng the average prce of real estate wthn t. Thus, one mportant task n modellng the mass apprasal of real estate s assessng the nfluence of an area on the real estate wthn t, as well as descrbng the gven area wth a suffcent number of attrbutes n order to determne the relaton between the values of these attrbutes and average property prces n each spatal unt. By descrbng the mddle set of attrbutes, and usng the known data on the average prces of property wthn t, a base of known cases can be formed knowledge that wll serve as a standard for prcng propertes n any area, whch can also be descrbed by means of an dentcal set of attrbutes. Usng logcal aggregaton or aggregate measures of smlarty, that s by determnng the measure of smlarty between the assessed spatal unts and representatves from the lst of known cases, the most lkely value of the average prce of real estate can be antcpated. Namely, we can consder an aggregate measure of smlarty as a pont n n-dmensonal real vector space, where the dstance of the rectangular proecton pont from the pont of orgn for each axs s equal to the value of a partal measure of smlarty for each attrbute. On the bass of the dstance from the pont of orgn, the overall measure of smlarty to the lst of known cases for the spatal unt concerned can be determned, and on the bass of the measure of smlarty, the average prce of property wthn t can be antcpated.

3 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, Based on analyss of the real estate market, on the natonal spatal data nfrastructure and on relevant data from other sources, t s possble on one hand to assess the value of some spatal unts for all of the defned attrbutes, and on the other hand on the bass of the market to defne the average prce of property n t. On the foundaton of ths data a database of knowledge can be formed, or a base of known cases whch can stll be used for antcpatng the prce of property n other spatal unts. Namely, by usng nterpolatve Boolean algebra and logcal aggregaton as aggregaton operators for measurng smlarty, one or more smlar spatal unts are found on the bass of the nput data values of all attrbutes, based on predefned crtera, the average prces of whch are used as a bass for antcpatng the most probable average prce of property n a gven spatal unt. Valuaton of property usually takes place n developed market economes, and Serba, realstcally, s only at the begnnng of ths. Methodologcal nconsstency, nsttutonal conflct and uncoordnated ursdcton are only some of the characterstcs whch nevtably mply the prevous statement for the test area. The feld of applcaton of the assessment results s a multdmensonal space usng ths nformaton mostly for economc purposes. Provdng fundng, nvestment decson makng, statstcal or fnancal reportng, makng busness decsons, legal practce, busness nsurance and mplementaton of fscal polcy are ust some of the areas for whch estmatng real estate value provdes data, on the bass of whch mportant and often dffcult decsons are made. The concept of market value s as dffcult to unambguously defne a show well the procedure for estmatng sutable real estate s performed. 2. Models of Mass Apprasal of Real Estate Value As wth most estmates that a man makes each day, so t s wth estmatng the value of real estate. Behnd the overall ratng whch s defned ether numercally or semantcally (lngustcally), n fact, hdes, a mult-crteral analyss and the optmzaton of dfferent crtera. Although the ndvdual s not usually aware of them, and although they would be hard to defne, the crtera taken nto account on that occason are not only numerous, but n ths thnkng process ther rankng s also carred out. However, an attempt to carry out an obectve estmate n one thnkng process on the bass of dfferent crtera would more often than not be futle wthout some of the known methods of operatonal research soft computng that s, artfcal ntellgence. The number of crtera on the bass of whch to descrbe or estmate a property s very large, and together they are ntended to represent ther constructonal, economc, socal, ecologcal and other aspects. We are contemporares of the fact that not only for the tme functon but also for the locaton, the evaluaton crtera are subect to certan modfcatons, that s, adaptng the basc requrements of the average buyer on the property market. It s not surprsng that buldngs whch are dentcal n terms of ther techncal constructon crtera have dfferent market values n dfferent locatons. There are also places or even regons whch, because of ther basc economc, demographc, socologcal and other characterstcs have dentcal real estate values for

4 42 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, smlar propertes. Lookng only through the prsm of estmaton, and frequency of operaton, supply and demand, and therefore the prce of real estate show that there are other relevant crtera that sgnfcantly nfluence the value, whch wll be dscussed n the followng consderaton. It s also mportant to emphasze the fact that certan features that affect property value n one place may be rrelevant n another place. For example, f a populated area s near a hghway, by means of whch t has a good connecton wth a maor natonal or nternatonal road corrdor, the value of land n ts vcnty drectly relates to ts dstance from that hghway, a potental factor whch can ustfy nvestment n t economcally. However, f there s no such traffc route n the vcnty of a populated area, then t s an rrelevant crteron for all property n that area. A smlar example can be gven for the water, ral, sport recreaton, spa, ar and educatonal facltes whch characterze or do not characterze a place Cost method One of the basc prncples of ths methodologcal approach s to determne the obectve prce, that s the cost of buldng the property, that s replacement theory. The logcal foundaton of ths approach s reflected n the statement that the unnformed buyer n the open real estate market could pay a hgher prce for a property than the actual cost of ts constructon. It can be seen mmedately that n ths methodologcal approach the tme factor s neglected, as well as many other components, some of whch can and others whch cannot be defned quantatvely, that s n terms of currency unts, and whch would defntely be used up durng ts constructon, and fctvely mss the proft from the eventual use of the exstng buldng. Bearng n mnd ts basc methodologcal approach, the cost method can be consdered as effectve when estmatng the value of newly constructed buldngs or buldngs whch are not affected by the tme factor, or for buldngs whose explotaton perod does not have to be consdered as a component that affects ts deprecaton. Ths method could also be consdered as sutable when estmatng property for whch t s dffcult to assess the beneft ganed from ts beng rented out, or for cases n whch the assessor does not have all the nformaton about the cost of smlar or dentcal propertes n the localty, that s, when t s mpossble from the base of known cases to fnd any knd of analogy wth the property n queston, and on the bass of that draw a concluson about ts value. The cost method s used for estmatng ndustral, agrcultural and other real estate, as well as beng the domnant and only method for state admnstraton procedures and court proceedngs (Mladnovæ 2009) Mass apprasal of real estate value Consstently followng the basc dea of establshng cadastral records and the correspondng requrements of modern socety where real estate s a very mportant aspect of everyday lfe, the modern cadastral records of developed countres across the whole world are ntroducng models of mass apprasal (mass estmaton) of real estate wthn the framework of ther everyday actvtes (Mladnovæ 2009).

5 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, Modern techncal and technologcal developments, the development of nformaton technology n the admnstraton of large amounts of data n real tme, modern methods of acqurng large amounts of good qualty and relevant geospatal data, operatonal research methods, artfcal ntellgence and multcrtera optmzaton have created the necessary precondtons for the mass apprasal of real estate value. Georeferencng a large amount of heterogeneous spatal data, ts statstcal qualty control, as well as relatonal searches wthn geonformatonal systems have only further accelerated and facltated the applcaton of mass apprasal of real estate, made opportuntes n the functonng of tax admnstraton and facltated ts connecton wth other publc bodes. The ntenton of mplementng of tax polcy n modern socety s that taxaton on the bass of property ownershp s based on determnng the actual, obectve market value n the shortest possble tme perod. One of the requrements s also for the best possble qualty control methodology on the bass of statstcal evaluatons, as well as the need for a flexble and unversal system, practcally applcable for the mplementaton of fscal polcy. Mass apprasal fnds ts role prmarly n formng ntal or reference values, whle the methodology of ndvdual estmates s used to survey the peculartes and characterstcs of a partcular property n order to defne ts real value, whch serves as the tax base. For qualty and obectve mplementaton of fscal polcy, n addton to the qualtatve methods of mass that s ndvdual estmaton, an ndspensble component s the admnstratve skll necessary for managng the human and physcal resources of the tax department and the qualty assurance at every level of the mass apprasal process. One of the fundamental dfferences between mass and ndvdual apprasal s that the mass model has to examne the wder context and defnes the characterstcs whch nfluence the formaton of real estate values over a larger area. Thus, apprasal refers to group rather than to ndvdual propertes, and the task of the assessor s to dentfy and recognze the common features of a large number of propertes, that s, he must be capable of developng, supportng and explanng standardzed adustments to the model of apprasal, as well as usng class, type of buldng, envronmental and other groups of real estate. The next mportant dfference between these two methods s that the qualty control of the apprasal results s carred out n dfferent ways. Namely, ndvdual estmates relate to a specfc property whch an ndvdual or a small group of ndvduals are nterested n, the result of whch can be compared wth the research or analyss for smlar, that s, comparable sales of smlar property. For mass apprasal, qualty testng s carred out usng statstcal methods. The subect of mass apprasal s a larger amount of real estate, and therefore the number of nterested clents s larger tax payers, who must on one hand be satsfed or at least unharmed, and on the other hand the model must be consstent and unversally applcable throughout the whole area n whch the tax polcy s mplemented under the auspces of an dentcal legal framework. Unlke ndvdual apprasal, t s almost mpossble n ths day and age to consder the methodology of mass apprasal wthout computer support. It not only revews

6 44 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, relevant crtera more obectvely n a wder context, but s also fast and effcent n ts applcaton and has effectve control over ts mplementaton as well as beng relable n the results of the model appled. The mass apprasal system, whether computerzed or manual, conssts of four subsystems: 1. Data management system 2. Sales analyss system 3. Apprasal system 4. Admnstratve system. The four subsystems are nterdependent. The apprasal system, for example, uses nformaton held n the sales analyss system, and the data management system produces output documents necessary to the admnstratve system for prntng tax blls. In Serba, the Law on State Survey and Cadastre envsages the ntroducton of mass apprasal of real estate, whch should be an extremely sgnfcant event n ths area and one whch wll gve the ntatve to regulate the other aspects of apprasal accordng to the regulatons. Tranng personnel and establshng an assocaton of apprasers and other nsttutons are tasks that le ahead on the road to systematc and nsttutonal regulaton of ths very mportant area. 3. Evaluaton of Spatal Unts based on CBR and LA One of the goals of the scentfc research n ths paper s the recognton of the relevant features, that s, crtera of spatal unts whch affect the formaton of the average prce of real estate wthn them. The ntensty or value of each of them speaks of the value of that localty from the aspect of the crteron n queston, and the ntegraton of all the ndvdual values gves the total value. A very mportant problem n determnng the level of smlarty s the aggregaton (fuson) of a number of attrbutes nto one globally representatve aspect the measure of smlarty. In exstng practce, the most commonly used method for summng up the weght coeffcents of partal aspects s the aggregaton technque. Ths approach s addtve, and for all cases whch are not addtve, t s nadequate. For example, usng a weghted sum for aggregaton, even n cases wth only 2 attrbutes (a, b), does not allow realzaton of the natural need for a requrement n whch both attrbutes are mportant together. Among those who study the mult-attrbute decson makng method, ths problem s known and as a soluton, the theory of capactance, known n the phase-communty as phase measures and phase ntegrals (Mrkovæ et al. 2006). In ths knd of approach, addtvty s a relaxed monotoncty for whch addtvty s a specal case. As a result, the range of possble applcatons for ths approach s consderably broader, however from a logcal pont of vew, monotoncty stll remans a strongly lmtng factor, snce many logcal functons by nature are not monotone. The generalzed dscrete Choquet ntegral s defned for a general measure non monotoncty n general. Ths approach ncludes all logcal and pseudo-logcal functons, but allows the use of only one arthmetc opera-

7 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, tor for nterpolaton purposes the mn functon. Interpolatve realzaton of Boolean algebra (IBA) ncludes all logcal functons and all nterpolatve operators (operators of a generalzed product) (Mrkovæ et al. 2006). The proposed model, therefore, on the bass of measures of smlarty between cases (spatal unts) for whch the average values of real estate are determned and representatves from the base of known cases antcpate the most lkely average prce. Measures of smlarty are a type of evaluaton functon whch transfers (maps) an abstract concept nto a numercal value.e. ons values together n a seres of pars, wth the dea that a hgher value ndcates greater smlarty, and that value s an aggregated, composte sze. Smlarty, as a complex quantty, can be expressed by means of unque values whch are reached by the aggregaton of partal measures of smlarty (for each of the selected attrbutes-crtera). The possbltes of the logcal aggregaton model extend the framework of the exstng model. In most well-known models, only trval attrbutes are used, wthout nteracton between them (wthout the use of logcal functons). Also, for the aggregaton operator, most commonly ust a normal product s used. The logcal aggregaton model s generalzed to allow the use of logcal functons between attrbutes (from a fnte set of possble functons for a fnte number of attrbutes), as well as the use of dfferent aggregaton operators and dfferent generalzed products subclasses of T-norms wth the full axom of non-negatvty (Radoevæ 2006). In addton, a measure of aggregaton does not have to be ether addtve or monotone. Determnng partal measures of smlarty s a trval task for the evaluator, snce there are references for comparsons performed n one-dmensonal space for the values of each ndvdual attrbute, so expert knowledge s not necessary n the way that t s for absolute (prmary) evaluaton. The mplementaton of logcal aggregaton s possble wthn the concept of CBR n the phase of aggregatng the data determnng the measure of smlarty Model based on the Concepts of CBR and LA Matrx O represents the knowledge base whch conssts of the spatal unts prototypes for whch the average property prces n them are known. Thus, the cases consst of a set of attrbutes whose values are nformaton carrers about the problem and correspondng average prces, and s the nformaton carrer regardng the soluton to the problem. Each prototype O represents a case a vector whch conssts of normalzed attrbute values a [ 01], and the average prce of real estate c. The case ndex s. The attrbutes are defned as: O a, ( 1, k) (1) For determnng the average prce n spatal unt O x t s necessary to normalze the attrbute values n the same way as was done n the knowledge base. Then the measure of smlarty s determned wth prototypes from the knowledge base.

8 46 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, The measure of smlarty s an aggregate sze. It s a result of the aggregaton of ndvdual (partal) smlarty measures for each attrbute separately. A partal measure of smlarty s actually a measure of logcal equvalence. An approprate aggregaton for the values of smlarty measures s: x x ( a a ) [( a a ) ( Ca Ca )] 1 a a 2a a x x x : mn (2) Snce ths s aggregaton of the same attrbute (hgh postve correlaton), for a generalzed product, we use the mn functon. The aggregaton operator carres out mappng: Aggr:[ 0, 1] 2 [ 0, 1] (3) The total smlarty measure s ganed by aggregatng partal (per attrbute) measures of smlarty: k k ( 1 ), : (4) 1 Snce ths s aggregaton of dfferent attrbutes (neglgble correlaton), for a generalzed product we use an ordnary sum. The aggregaton operator carres out mappng: Aggr:[ 0, 1] [ 0, 1] (5) There s a possblty of allocatng weght coeffcents w to partal smlarty measures. A relevant total measure of smlarty s then determned by the expresson: k w ; 1 k w 1 1, w 0, 1, k (6) There remans an open possblty of aggregatng a hybrd-mxed type as requred usng nterpolatve pseudo-boolean polynomals. The most smlar prototype O O wth O x s the one wth the hghest value of smlarty measure : Oxc O, O O where s chosen for max (7) In ths way, the value of the average property prce s taken for a spatal unt (case) from the knowledge base whch s most smlar to the nput case, for whch we want to fnd the average prce. Then these prce values are assgned to the case, the average prce of whch s requred The mathematcal bass for the suggested model of evaluatng spatal unts The mathematcal model suggested on the bass of scentfc research wthn the framework of ths doctoral dssertaton s based on the prncple of case-based reasonng n whch a combnaton of logc and the Eucldean (L2) norm are used.

9 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, Namely, antcpaton of the average prce values of real estate wthn the framework of a spatal unt s realzed on the bass of aggregate measures of smlarty between the spatal unt for whch t s carred out and representatves from the base of known cases knowledge. Aggregate smlarty measures are the ntegraton of ndvdual of partal measures of smlarty for each of the attrbutes that descrbe all of the cases, and n ths case a partal smlarty measure s a measure of logcal equvalence. As ndcated above, an approprate aggregaton operator for the value of smlarty measures s: x x x ( a a ) [( a a ) ( Ca Ca )] 1 a a 2a a ; : mn (8) The total the aggregated measure of smlarty can be represented as pont ( O ) n n-dmensonal real vector space, where the dstance of the rectangular proecton pont from the pont of orgn for each axs s equal to the value of partal smlarty measures for each of the attrbutes. Usng the Eucldean norm as an aggregaton operator for measures of smlarty actually calculates the measure of dstance of pont ( O ) from the pont of orgn as follows: r 2 k 1 1 x 2 where k s the number of attrbutes (9) x In the proposed model there also remans the possblty of allocatng weght coeffcents to each of the squares of partal measures of smlarty. Then the dstance s defned wth: r k 2 w 1 1 k 2 ; where: w 1, w 0, 1, k (10) Then the value of the total smlarty measure s equal to the value of the logcal negaton of the dstance. 1 r r ( ) 1 (11) The most smlar prototype O O wth O x s the one wth the greatest value of smlarty measure, whle the greatest value of smlarty measure takes the case where the dstance from the ponts r s the least. Under the proposed model for every new case (spatal unt) whch s valued, from the knowledge base an already known case s found to whch t s most smlar. As descrbed above, the smlarty measures wll be normed values from the nterval [0.1]. However, the queston s whch measure of smlarty s a border case for whch t can be sad that a known case from the knowledge base can be used to antcpate the average prce values of real estate for the case beng assessed. Wthn the framework of the proposed model, n ths doctoral dssertaton, a mnmal value of smlarty measures of 0.8 s adopted. Accordng to ths, the prncple s kept by whch, on the bass of total measures of smlarty, t s not necessary to fnd the most smlar case, but rather n determnng the average prce value of real estate wthn the frame-

10 48 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, work of evaluatng spatal unts, all known cases are used whch, when evaluated, have a total smlarty measure greater or equal to the gven border values. The estmated average value of real estate prces wthn the framework of the assessed spatal unts s not only a smple arthmetc mean of the average prces for known cases from the knowledge base whose measure of smlarty wth assessment are greater than or equal to the border values, but rather, t s proportonal to t. The antcpated value s acheved usng the followng formula: n ( 08. ) PRICE prce 1 ; where ( 08. ) n 1 (12) n the number of cases from the knowledge base whose total measure of smlarty to the case under assessment s greater than or equal to the border value. 4. Quanttatve Defnton of Parameters In any procedure of mult-crtera evaluaton, t s necessary, above all, to defne the crtera on the bass of whch the procedure wll be carred out. Each of the crtera can have approprate values whether they are expressed numercally or semantcally. The value of each crteron represents a score, n ths case, each spatal unt from the aspect of the crtera used to evaluate t. In order for the process of evaluaton, that s, for the assessment of the value of each ndvdual crteron to be mplemented, t s necessary for each of them to be descrbed, that s, to defne the method for ther evaluaton. The crtera used n ths scentfc research are selected on the bass of two basc postulates: to better descrbe the spatal unts beng assessed from the vewpont of the real estate value wthn them, and for the data used to evaluate the attrbutes to be avalable from a relable source so that when appled to future models, the same source can be used for the actualzaton and modfcaton of the suggested model. In creatng a model of mass apprasal, the crtera for evaluatng spatal unts, whch, accordng to ther nature, are classfed nto four basc groups, whch are: natural, socal and economc characterstcs, or crtera whch can be consdered as a correctve factor n mass apprasal. The basc concept of establshng ths model s that on the bass of the values of all crtera for each spatal unt, as a result of the estmaton, a characterstc number normalzed from 0 to 1 s acheved, whch s bascally a value crteron, that s, a ranked spatal unt whch should correlate wth the average value of the real estate wthn t. Each group of characterstcs n tself has more ndvdual crtera or sub-crtera, the value of whch s also normalzed wthn the framework of each group n ntervals from 0 to 1. In other words, the evaluaton s decomposed nto several layers, and for each of these layers, normalzaton s carred out usng a bottom up prncple, untl one spatal unt reaches a characterstc number. For easer and more systematc presentaton below, the sub-crtera used and the method of ther normalzaton n ndvdual groups are descrbed frst, followed by

11 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, a presentaton of the normalzaton of a characterstc number of spatal unts as a result of mass apprasal Natural Characterstcs Among natural characterstcs are those whch n any envronment, a man s lfe and work should not be able to nfluence. For the purposes of the scentfc research n ths paper, the followng natural characterstcs are used as crtera for the evaluaton of spatal unts: 1.1. The ecologcal aspect 1.2. Dstance from the captal cty 1.3. Geostrategc locaton Transport corrdors Borderng countres Natural resources. By consderng these crtera n a slghtly broader context, t s possble to come nto conflct wth the prevous paragraph by statng that actually, man does ndrectly nfluence these factors. We are contemporares of the growng mpact man has on ecology, whch s unfortunately manly negatve. The expanson of ctes and mgraton from smaller communtes to large urban centres nevtably leads to huge expanson and over tme, a change n the dstance between large centres and smaller communtes must be consdered. If n ths regard, techncal and technologcal progress s taken nto account, then that physcal dstance or dfference takes on another dmenson. We are also contemporares of the fact that at dfferent tme ntervals the concept of geostrategc poston has often changed n our regon, and therefore the mportance of ts nfluence through hstory has had dfferent ntenstes. The result of the assessment, that s, evaluaton of spatal unts wth respect to ther natural features s a real number a normalzed value whch takes a dscrete value n the nterval from 0 to 1. A dscrete value s obtaned by breakng down the evaluaton crteron from the perspectve of natural features nto sub-crtera, as prevously shown. Each of these sub-crtera (ecologcal aspect, dstance from the captal cty and geostrategc poston), also as a result of the evaluaton of the local envronment, has as a resultng normalzed value n the nterval from 0 to 1. Hence, the maxmum value of each of these three sub-crtera s 1. The resultant from these three sub-crtera s calculated as the sum of these three values whch are assgned approprate weght (coeffcents) as shown n the Table 1. Table 1. Decomposton of the crteron natural characterstcs. Crteron Sub-crteron Value Coef. Norm. Value. 1. ecologcal aspect [0,1] 0.15 [0,0.15] 1. Natural characterstcs 2. dstance from the captal cty [0,1] 0.5 [0,0.5] 3. geostrategc locaton [0,1] 0.35 [0,0.35] Sum=1 Sum max =1

12 50 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, Socal Characterstcs The socal characterstcs of any spatal unt, as opposed to the natural ones, nclude all of those trats, that s, characterstcs whch greatly depend on the populaton and ther actvtes n that area. For the purposes of the scentfc research n ths paper t was necessary to dstngush those characterstcs or crtera whch, from the socal aspect of a spatal unt, affect the value of the real estate n t. In ths respect the crtera used to estmate or evaluate a spatal unt n the suggested model are: 2.1. Populaton densty 2.2. Increase n the number of nhabtants 2.3. The number of employed 2.4. The educatonal aspect. The evaluaton value for spatal unts from the aspect of socal characterstcs or crtera s a normalzed value a real number from the nterval from 0 to 1, whch s obtaned as the sum of the normalzed values of the sub-crtera broken down n the method prevously shown. Decomposton of socal characterstcs (crtera) nto sub-crtera: populaton densty, ncrease n the number of nhabtants, the number of employed and the educatonal aspect has the am of evaluatng a spatal unt n the best way possble to gve the most relable evaluaton that s to obtan the most obectve evaluaton possble of ther nfluence on real estate value. The normalzed value of each sub-crteron s gven a correspondng weght whch defnes the nfluence or mportance of each sub-crteron wthn the total assessment of the crteron n queston. The method of decomposton and evaluaton of the crteron socal characterstcs s shown n the followng Table 2. Table 2. Decomposton of the crteron socal characterstcs. Crteron Sub-crteron Value Coef. Norm. Value. 1. populaton densty [0,1] 0.15 [0,0.15] 2. Socal characterstcs 2. nc. n no of nhabtants [0,1] 0.05 [0,0.05] 3. number employed [0,1] 0.4 [0,0.4] 4. educatonal aspect [0,1] 0.4 [0,0.4] Sum=1 Sum max = Industral Characterstcs On smlar way as prevously presented n the procedure of evaluatng spatal unt t could be analysed ndustral characterstcs as well. The method of decomposton and evaluaton of the crteron ndustral characterstcs s shown n the followng Table 3.

13 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, Table 3. Decomposton of the crteron ndustral characterstcs. Crteron Sub-crteron Value Coef. Norm. Value. 1. average earnngs [0,1] 0.4 [0,0.4] 2. agrcultural development [0,1] 0.1 [0,0.1] 3. Industral 3. GDP [0,1] 0.2 [0,0.2] 4. road nfrastructure [0,1] 0.1 [0,0.1] 5. toursm [0,1] 0.2 [0,0.2] Sum=1 Sum max =1 5. The Results of Testng the Suggested Model Wthn the framework of the scentfc research n ths paper, the suggested model for evaluatng spatal unts was tested on a sample of 30 towns equally spaced over the whole test area. The data used n testng the applcaton of the model was taken manly from offcal statstcal data from the Serban Statstcal Offce, or from other offcal publc sources. It should also be noted that testng was not based on any data from 2009 or The reason for ths fact s the greatest world economc crss snce the Second World War, as a consequence of whch were large fluctuatons n the real estate market, as well as sgnfcant oscllatons n the prces of the same. Sgnfcant prce fluctuatons are a drect consequence of ths crss, and not any other reason, whether economc, commercal or related to any other characterstc, and whch consderably nfluence the market prce. Another mportant reason les n the fact that ths perod concdes wth the adopton of a new plannng and constructon law n Serba, whch ntroduces sgnfcant changes wthn ths feld. One of the consequences of these changes s a pronounced decrease n the dynamc of constructng new resdental and commercal buldngs, partly due to slow and poor mplementaton of the law, and the non-exstence of secondary legslaton, and partly because of the mentoned economc crss. Also, one of the poltcal decsons of the government of the Republc of Serba, that by stmulatng constructon and ntensve buldng of socal housng, the prces of whch wll be sgnfcantly lower than the market value, t wll revve ts own economy n a perod of world economc crss. Ths decson sgnfcantly affects the creaton of real estate market prces, whch can also not be treated as usual market characterstcs. All of the above mples that usng data from the real estate market from the past two years would call nto queston the qualty of the model. At the same tme t would burden the results wth ts use, as t relates to a very specfc non-standard perod whch sgnfcantly devates from the legal prncples whch would otherwse apply under normal crcumstances. Below are the clearly presented results of the valuaton of spatal unts, as well as the antcpated average values of a square metre of real estate n them. The prncple establshed for the purpose of ths scentfc research s that on the bass of the lst of known cases comprsng 29 places descrbed by a set of attrbutes, the prce of a spatal unt beng valued s based on the values of aggregate smlarty mea-

14 52 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, sures between the spatal unts n queston and correspondng representatves from the lst of known cases. Vsualzaton of the results acheved was carred out on thematc maps of the test area, and s presented n concse tables. For the purpose of presentaton the results of ths nvestgaton n the paper the results of evaluatng of the spatal unt Savsk Venac, n the Table 4 and correspondng Fg. 1. Table 4. The results of evaluatng of the spatal unt Savsk Venac. No. Town Aggregate measures of smlarty Natural Characterstcs Socal Characterstcs Socal Characterstcs DEFINI TELY Average prces -0.8 Prce 1. Nov Beograd Zemun Savsk Venac Pallula Subotca Zrenann Kknda Panèevo Apatn Baèka Palanka Nov Sad Ruma Peænc Šabac Ub Smederevo Velko Gradšte Aranðelovac Jagodna Bor Sokobana Bana Bašta Ivanca Raška Kruševac Nš Blace Prot Medveða Vrane =

15 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, Fg. 1. The thematc map of the results of evaluatng of the spatal unt Savsk Venac.

16 54 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, In Table 5 s a comparatve presentaton of the average prce values n the spatal unts where the suggested model was tested. Also presented s a comparson of the average market prces and the prce obtaned by antcpaton or by applyng the suggested mathematcal model, n whch the percentage rato of the same s gven. Table 5. The results of testng of the proposed model of evaluatng spatal unts. No. Town Average prces Values from the model % Increment the prce n % 1. Nov Beograd Zemun Savsk Venac Pallula Subotca Zrenann Kknda Panèevo Apatn Baèka Palanka Nov Sad Ruma Peænc Šabac Ub Smederevo Velko Gradšte Aranðelovac Jagodna Bor Sokobana Bana Bašta Ivanca Raška Kruševac Nš Blace Prot Medveða Vrane

17 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, As s evdent from the above table, the proposed model of evaluatng spatal unts, based on the concept of case-based reasonng usng nterpolatve Boolean algebra and logcal aggregaton, confrms the scentfc hypothess that the proposed model enables the antcpaton of the average prce of real estate n the test area whch s n accordance wth real (market) prces wth a mnmum amount of 75 to 80%, and t shows a very hgh level of agreement between market and antcpated prces obtaned from the model. By thorough analyss of the presented results, t can be concluded that n ndvdual cases, the suggested model gves results whch devate sgnfcantly from the real (market) values, whch wll be cause for further explanaton of these anomales. Savsk Venac. Namely, one of the extremes whch reflects the sgnfcant devatons s the example of the Savsk Venac Muncpalty, whose antcpated average property value of 1295 makes up only 55% of the real market value whch amounts to 2347 euros. There are a number of real reasons for ths knd of anomaly. One of these certanly les n the fact that Savsk Venac s the muncpalty wth the hghest average property prce n the test area, wth prces sgnfcantly hgher than any n the test area. Because the proposed model s desgned to antcpate a value based on measures of smlarty wth known cases from the knowledge base, that s, based on the value of the same cases, t s natural to expect that such anomales arse. In other words, a knowledge base should be created so that each new unknown case s resolved on the prncple of nterpolaton, snce t s natural that n ths type of model, the prncple of extrapolaton does not result n the desred outcome. On the other hand, the specfcty of Savsk Venac, and other Belgrade muncpaltes, les n the fact that the admnstratve dvson of Belgrade nto muncpaltes cannot be treated n the same way as other cases n the test area. The dvson nto muncpaltes, n ths case more than n other cases, represents an admnstratve boundary, but not a boundary that represents a change n ambent, condtons, qualty, comfort, culture and lfestyle. By descrbng and consstently evaluatng urban muncpaltes wth a set of defned attrbutes and havng n mnd ther terrtoral ursdcton, t s very easy to arrve at sgnfcant errors n the applcaton of the model. Such consstent applcaton of the evaluaton of urban muncpaltes would gnore the fact that, for example, an mportant nternatonal traffc route whch passes through the centre of Belgrade represents equal qualty for all of the muncpaltes, although admnstratvely speakng, that same traffc route passes through only a few of them. We can also observe the example of rvers and streams. The Sava and Danube rvers offer a great qualty of lfe n Belgrade, and regardless of ther admnstratve belongng to partcular muncpaltes they represent not only the prde but the value of the whole cty. Also, no person would say to the resdents of a partcular muncpalty that he lves n a place that s not a unversty centre, even though the whole unversty nfrastructure s located n ust a few muncpaltes. Smlar examples can be gven for cultural hstorcal monuments, mportant hotel accommodaton, cultural, admnstratve, relgous, sport and other urban centres. If we added to these facts the specfctes of muncpaltes t could mean,

18 56 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, for example, that an mportant unversty centre n one muncpalty n a street whch s on ts admnstratve border, s opposte the street to buldngs that are n another muncpalty whch does not have such an nsttuton. By consstently applyng the model, buldngs stuated several klometres from an mportant unversty centre, yet belong to that muncpalty, wll have the same value as buldngs whch have that knd of centre n ther muncpalty, and those who admnstratvely do not belong to that muncpalty, but are only a few steps from the mportant centre, wll not have that qualty, whch s a more than obvous anomaly whch should be taken nto account durng the evaluaton process. The example of Savsk Venac hdes a fact that s very mportant to take care of and have n mnd durng valuaton. Namely, t s the muncpalty where the most attractve locatons are stuated, not only n Belgrade but n the whole of the test area, whose average property prces do not represent a quanttatve manfestaton of the qualty of that property, but rather that the attractveness of the locaton s more mportant than the place of lvng. Ownershp of property n these locatons speaks of the fnancal, socologcal, cvl or other status of the property owner, although by evaluatng these locatons accordng to a set of proposed attrbutes, the results would not reflect ths. For ths knd of locaton, an addtonal qualty s the absence of everythng whch n any other place would make ts qualty. By analyzng the fact that n these locatons there s no maor nfrastructure, or sgnfcant economc, admnstratve, retal, unversty, cultural or any other centre, t s possble to arrve at errors related to the value of real estate n them. However, t s only n large urban centres lke Belgrade n whch all that s mssng n that locaton can stll be found n the vcnty, that such absence can be referred to as a qualty, snce f all of there attrbutes could not be found wthn a reasonable dstance, t would certanly not be consdered as a qualty. Ths can be seen n the example of some naturally more attractve locatons on the perphery of Belgrade, where property prces are sgnfcantly lower and the locaton s not nearly as attractve. Bearng all of ths n mnd n a model, partcular attenton should be pad to urban centres so as not to mslead a user nto usng the model lterally, takng nto account only the admnstratve and terrtoral aspects of muncpaltes. 6. Conclusons The proposed model s tested on a lmted sample, wth ts results confrmng a hgh level of correlaton between the market and antcpated average prce values of real estate n the subect spatal unts. Also, the tested model ndcates the need for certan specfcs of a local area to be taken nto partcular consderaton, especally when some of the attrbute values are extreme, and as such domnant n the formaton of the value of real estate wthn that area. Ths model s a good bass for solvng the comprehensve problem of mass apprasal of real estate values, and ts mathematcal base provdes a consstent approach to problem solvng at all levels and should therefore be modelled and used

19 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, at all levels n the formaton of a comprehensve model of the mass apprasal of real estate values. Testng the model confrmed the hypothess that t s possble to conceve a model that, for the gven nput values of the attrbutes of spatal unts, fnds the most smlar from the knowledgebase prototypes, and on the bass of them assesses the most probable average value. If t does not fnd a value wth the gven measure of smlarty, t s capable of expandng the knowledge base wth the gven case, so that over tme, that base would become better and more relable. A model conceptualzed n ths way s consstent n everythng wth the basc assumptons of ths scentfc research and t should serve as a startng pont n the creaton of a model of mass apprasal of real estate n the Republc of Serba, and as such t could be unversally applcable, not only n the test area, but also n a wder context. Ths model, lke any other should be perodcally revewed, that s, t must comply wth all relevant factors affectng the market value of real estate ncludng economc, commercal, socologcal, ecologcal, securty and any other factors. Hence, the choce of relevant attrbutes n all models, takng nto account the dfferences both n dependence on the level of generalty and n the classfcaton of real estate, should be consdered n a tme functon. The proposed concept of an operatve model provdes flexblty and the possblty of recognzng changes n relevant crtera values that nfluence the value of real estate at a gven moment. ACKNOWLEDGEMENTS. The work reported n ths paper s a part of the nvestgaton wthn the research proect TR supported by the Mnstry for Scence and Technology, Republc of Serba. Ths support s gratefully acknowledged. References ASA (2008): Busness Valuaton Standards, Amercan Socety of Apprasers, Herndon. Delbašæ, B. (2007): Formalzng the process of busness decson makng, doctoral dssertaton, The Faculty of Organzatonal Scences, Belgrade Unversty, Belgrade, (n Serban). Kuburæ, M., Ærovæ, G. (2011): Model of valuaton of spatal unts based on case based reasonng as a bass for mass apprasal of real property value, Proceedngs of paper, 1 st Serban Geodetc Congress, Belgrade, Law on the establshment of cadastral ncome, RS Offcal Gazette, No 49/92, (n Serban). Law on Property Taxes, RS Offcal Gazette, Nos 26/01, 80/02, 135/04 and 61/07, (n Serban). Law on State Survey and Cadastre, RS Offcal Gazette, No 72/09, (n Serban). Law on Mnstres, RS Offcal Gazette, Nos 65/08 and 16/11, (n Serban).

20 58 Kuburæ, M. Ærovæ, G.: The Applcaton of Intellgent Technques, Geod. lst 2012, 1, Mladnovæ, M. (2009a): Valuaton of Real Estate, Faculty of Cvl Engneerng, Belgrade Unversty, Belgrade, (n Serban). Mladnovæ, M. (2009b): Valuaton of Real Estate, Proceedngs of the fourth summer school of urban plannng, Assocaton of Town Planners of Serba, Belgrade, Zlatbor, (n Serban). Mrkovæ, M., Hodolè, J., Radoevæ, D. (2006): Aggregaton for Qualty Management, Yugoslav Journal of Operatons Research 16, Number 2, Radoevæ, D. (2006): Boolean frame adequate for treatment of gradaton or fuzzness equally as for two-valued or classcal case, SISY 2006, 4th Serban Hungaran Jont Symposum on Intellgent Systems. Radoevæ, D. (2007): Logcal Aggregaton Based on Interpolatve Realzaton of Boolean Algebra, Eusflat Conf. (1), Saaty, T. L. (1972): An Egenvalue Allocaton Model for Prorzaton and Plannng, Energy Management and Polcy Center, Unversty of Pennsylvana. Saaty, T. L. (1977): A Scalng Method for prortes n Herarchcal Structures, Journal of Mathematcal Psychology, 15 (3), Watson, I., Marr, F. (1994): Case-Based Resonng: A Revew, The Knowledge Engneerng Revew, Vol. 9, No. 4, Prmena ntelgentnh tehnka za masovnu procenu nekretnna SAETAK. U radu e razmatran poam masovne procene vrednost nekretnna u okvru koega se, osm defnrana osnovnh pomova, zvod paralelna analza razlèth meðunarodnh skustava vezanh uz ovu problematku. Takoðer, analzra se normatvna nsttuconalna ureðenost predmetne celne na testnom podruèu, al se defnra poam vrednovana prostornh ednca kao osnove masovne procene vrednost nekretnna pole negove prmene. Defnrana e osnova modela vrednovana prostornh ednca temelenog na zakluèvanu po prncpu sluèaa (ZOS) logèke agregace (LA) te e dana matematèka osnova predloenog modela za predvðane proseènh cena nekretnna u okvru prostornh ednca. U predloenom modelu opsane su prostorne ednce, odnosno naèn nhove normalzace te nhova granulaca po grupama. Poednm atrbutma grupama dodelene su odgovarauæe tene koma se defnrau nhov poednaèn grupn znaèa u okvru ntegralnog modela. Na krau e zvedeno testrane modela na testnom podruèu. Kluène reè: vrednovane prostornh ednca, masovna procena vrednost nekretnna, zakluèvana po prncpu sluèaa, logèka agregaca. Prmleno: Prhvaæeno: