СLARA EXPERT VERBAL METHOD OF DETERMINING INVESTMENT RISK IN CONSTRUCTION

СLARA EXPERT VERBAL METHOD OF DETERMINING INVESTMENT RISK IN CONSTRUCTION Leonas Ustinovičius 1, Dmitry Kochin 2, Galina Ševčenko 1, 1 Department of Construction Technology and Management, Vilnius Gediminas Technical University, Saulėtekio al. 11, LT-10223 Vilnius, Lithuania E-mail: leonasu@st.vgtu.lt, galina@st.vgtu.lt 2 Institute for System Analysis, 117312, prosp. 60-letija Octjabrja, 9, Moscow, Russia E-mail:dco@mail.ru Abstract. Risk management is a systematic process for integrating professional judgments about relevant risk factors, their relative significance and probable adverse conditions and/or events leading to identification of auditable activities. In practical use task of getting expert knowledge in many cases can be formulated like task of classification, because expert intelligence is to sort objects (alternatives, state of object) through classes of decision. For example engineer analyzing breakdown in sophisticated technical system defines possible type of failure. Elements formatting some whole to be classified may have different origin. It can be different physical objects, cases of choice or condition of some object. The paper aims to presents a verbal method of determining investment risk in construction. The problem under consideration is investment assessment depending on the risk level. This article presents a new way to solve the problem - the CLARA expert verbal method.formally the problem is stated as one of multicriteria classification. A hierarchical approach to considered effectiveness indicators is proposed. The proof of the method effectiveness is presented. Keywords: investment risk, verbal methods, multiple criteria, non-structured problems, methods of solving multicriteria, classification problems, CLARA. 1. Introduction Risk management is a systematic process for integrating professional judgments about relevant risk factors, their relative significance and probable adverse conditions and/or events leading to identification of auditable activities [1]. In practical use task of getting expert knowledge in many cases can be formulated like task of classification, because expert intelligence is to sort objects (alternatives, state of object) through classes of decision. For example engineer analyzing breakdown in sophisticated technical system defines possible type of failure. Elements formatting some whole to be classified may have different origin. It can be different physical objects, cases of choice or condition of some object. Describing the method of prescription of the object to a certain class of decision is complicated because of inverbality of the strategy expert uses. Anyway these inverbal skills are effectively and promptly used, when expert solves task of classification in his sphere of knowledge. Classification is a very important aspect in decision making. One of the tasks preparing base for classification is setting of numerous criteria (attributes), which are able to describe any object. Scale of all criterion is formed by setting finite set of possible values. If in certain task scale of values of one or more criteria is infinite, it can be modified to finite by cutting it to finite set of intervals. Finally, on the base of expert knowledge must be organized classification of definite intervals and its components i.e. must be formulated rules according which any object can be prescribed to one of the predefined classes. Classified projects are described by assessing various efficiency criteria that could be both qualitatively and quantitatively expressed. Efficient project management is about risk management [6]. Risk management at the construction stage of the project life cycle influences not only the economic effect of the investment but also future efficient functioning of the building [12]. Lack of proper actions during construction can increase the probability of defects, failures, accidents or even catastrophes occurring. The purpose of this article is to demonstrate how multiple criteria can be used in analysis of facility

location problems. The article begins with an overview, explains the internationally most popular multiple objective analysis methods (ORKLASS, DIFKLASS, CIKL etc.), and demonstrates their applications on real problems. In multicriteria environment it is hardly possible to achieve this without resorting to special techniques. This article presents a new way to solve the problem - the CLARA expert verbal method. Formally the problem is stated as multicriteria classification one. A hierarchical approach to considered effectiveness indicators is proposed. The proof of effectiveness of the method is presented. A procedure of method application is described for a practical task. 2. Formal problem statement. Very often investment decision-making and research planning are referred to non-structured problems. Nonstructured problems have the following common characteristics. They are unique decision-making problems, i.e. every time a decision-maker is faced with an unknown problem or the one having new features compared with the previously considered case. Since the essential characteristics of such problems are qualitative, they can hardly be used in the analysis. On the other hand, the quantitative models are not sufficiently reliable. Given: 1. G is a feature, corresponding to the target criterion (e.g. building value). 2. K = { K 1, K2,..., K N } is a set of criteria, used to assess each alternative (course of treatment). q q 3. S q = { k 1,..., k } for q=1,..., N is a set of verbal wq estimates on the scale of criterion К q, w q is the number of estimates for criterion K q ; estimates in S q are ordered based on increasing intensity of feature G; 4. Y = S 1... SN is a space of the alternative features to be classified. Each alternative is described by a set of estimates obtained by using criteria K 1,...,KN and can be presented as a vector y Y, where ( y y ) y = 1, 2,..., y N, y q is an index of estimate from set S q. 5. C = { C1,..., CM } is a set of decision classes, ordered based on the increasing intensity of feature G. A binary relation of strict dominance is introduced as follows: ( x, y) {( x, y) Y Y q =1 N q y q Goal: on the basis of DM preferences to create imaginary F: Y { Yi }, i = 1,..., M, where Y i is a set of vector estimations belonging to class С i, satisfying the condition of consistency: x, y Y : x Y, y Y, x, y P i. i j ( ) j In this chapter some most frequently used verbal ordinal classification methods are considered. All these methods belong to Verbal Decision Analysis group and have the following common features [2, 10]: 1. Attribute scale is based on verbal description not changed in the process of solution, when verbal evaluation is not converted into the numerical form or score. 2. An interactive classification procedure is performed in steps, where the DM is offered an object of analysis (a course of treatment). A project is presented as a small set of rankings. The DM is familiar with this type of description, therefore he/she can make the classification based on his/her expertise and intuition. 3. When the DM has decided to refer a project to a particular class, the decisions are ranked on the dominance basis. This provides the information about other classes of projects related with it by the relationship of dominance. Thus, an indirect classification of all the projects can be made based on a single decision of the DM. 4. A set of projects dominating over a considered project are referred to as domination cone. A great number of projects have been classified many times. This ensures error free classification. If the DM makes an error, violating this principle, he/she is shown the conflicting decision on the screen and is prompted to adjust it. 5. In general, a comprehensive classification may be obtained for various numbers of the DM decisions and phases in an interactive operation. The efficiency of multicriteria classification technique is determined based on the number of questions to DM needed to make the classification. This approach is justified because it takes into consideration the cost of the DM s time and the need for minimizing classification expenses. Many different methods for solving multicriteria classification problems are widely known are presented in Table 1 [2, 6, 7, 8, 9]. ORCLASS [6], as an ordinary classification, was one of the first methods designed to solve these kinds of problems. More recent methods, such as DIFCLASS, CLARA and CYCLE have been applied for multicriteria expert analysis [7, 8, 9]. (1) Y Y q = 1... N P =. The algorithm CLARA (Classification of Real xq yq q0 : xq > y 0 q0 Alternatives) is based on the dichotomy of the One can see that this relation is anti-reflexive, antisymmetric alternatives chains, beginning with the longest chain. and transitive. It may be also useful to This concept, first used in the algorithm DIFCLASS [7] consider a reflexive, anti-symmetric, transitive binary and then in CLANSH [13], has been adapted for rarefied relation of weak dominance Q: spaces Y. Moreover, the algorithm CLARA uses a new Q =..., x }. idea of the adaptive dichotomy allowing us to (2) determine the boundaries between classes of solutions and make classifications much faster.

Table 1. Verbal analysis methods. The title of method The purport of method Notes ORCLASS DIFCLASS CYCLE CLANSH STEPCLASS CLARA This method is used for classifying different types of lendings.[6,7,8] This method was the first to use dynamic construction of chains covering alternative space for selecting questions to DM (decision men).[7,9] Algorithm CYCLE makes it possible to effectively build the complete noncontradictory bases of expert knowledge for the subject areas by complete order of the scales of criteria. [8, 9] Method CLANSH makes it possible to build the wheel bases of expert knowledge, when assumption about the presence of linear order of many estimations according to each of the criteria is substituted by assumption about the presence of the incoherent transitive binary relation.[8,13] System realizes technological approach to the structurization subject area and to the development of the decisive rules of expert and guarantees completeness and consistency.[9] CLARA has a much wider field of application, which includes the possibility of work with the scales of the criteria of arbitrary length with the partial order on the scales, and also in strongly rarefied spaces.[9] By deficiency algorithm appears its the large number of questions for DM to do the classifications. [6] The area of DIFCLASS application is restricted to tasks with binary criteria scales and two decision classes.[7] The methods can be successfully applied to classify investment projects when the decision classes and the criteria used are thoroughly revised. [9] 3. The main stages of gaining expert knowledge by using the method CLARA. The knowledge is gained by carrying on a dialogue with an expert. First, the main operations are outlined: 1. Discussing the statement of the problem. Defining the properties of G. 2. Generating a set of criteria K by an expert. 3. Constructing the scales for criteria evaluation. Preliminary analysis: checking if the estimates are (partially) arranged in the descending order of the distinctness of the property G. 4. Defining a set of ordered classes of solutions C by an expert. The second stage of applying the method expertmade classification involves submitting to an expert possible combinations of the attribute values for analysis. This is a time-consuming procedure because the number of combinations is usually large. This may entail expert s errors. Therefore, the method provides for defining some simple problems within the original classification problem by considering only two values of any attribute. Then, the results obtained are included in the original problem, and the expert solves this partially solved problem on the full scale. In the process of classification it may become clear that some combinations of the criteria values cannot be practically realized. In this case, the objects to which they refer are excluded from the analysis by an expert. The classification is over, when all the objects included in the analysis (a set Y*) are referred to a particular classes. At the third stage of analysis, the boundaries of classes are checked up because an expert could make some errors. Since class boundaries are the key factors in making classifications, every class specified by the expert should be checked up. For this purpose, every boundary element is offered to the expert again for checking. At the fourth stage, the defined boundaries of the classes converted into the expert rules of solution of the form: ab*** +, except {abcde,, abpqr}, so that every alternative should follow one rule, where ab*** is a fixed part of the rule, while is a rearrangeable part of the rule. Here, n is equal to the number of asterisks, ki is the number of estimates xi involved in rearrangement. The third part is activated if a set of alternatives described by a template is not completely rearrangeable, but, to achieve this, a small

number of elements is needed. Then, the missing elements are simply listed. The rules described are introduced into the system when solving the problem on a large scale and simplify the solution considerably by reducing the classification space. The decision rules of a particular class can be represented as a two-level tree (Fig. 1): Fig. 2. Risk evaluation Fig. 1. Decision rules of particular class where the values of key attributes are found at the higher level, while the combinations of the values of secondary attributes are found at the lower level. The rules described comply with inexplicit expert knowledge. The rules are submitted to an expert for approval. Some rules may be too complicated. In this case, the procedure of identifying a zone of superficial knowledge might be needed because complicated rules often indicate that knowledge is not stable. For this purpose, it is necessary to go back to the second stage of method application [11]. 4. CLARA (Classification of real alternatives). A classification problem, composed of risk evaluation criteria and final class decisions, is compiled for establishing investment project risks [11]. Constructional investment project risk evaluation criteria are provided in the first and second criteria levels. First hierarchy level the main one. Constructional investment project risk can be evaluated according to the criteria of this level. Each first hierarchy level criterion is allotted an evaluation: low, average, high or very high. When the evaluations are introduced, the result is received, i. e. risk levels are established. These criteria (I level) are not always enough for establishment of constructional investment project risk level. Therefore, each first hierarchy level criterion is split into lower level criteria. The second hierarchy level is composed this way. Criteria of the second hierarchy level are necessary for performing an accurate analysis (each of the risk types is analysed). Such risk evaluation work course is received following the drawn scheme (Fig. 2). Risk level might be established using the composed classificator, but a lot of criteria must be compared. It is a very difficult task for any person, besides it takes a lot of time. Therefore, it is possible to use computer program CLARA (classification of real alternatives). This method allows evaluating constructional investment project according to accurately established classes with the respectful offered criteria for risk size evaluation. 5. Risk determination and description of the operational factors of investment project. After a few iteration series such final decisions were chosen: 1. a) the lowest risk level; b) low risk level. 2. a) satisfactory risk level; b) average risk level. 3. a) high risk level; b) highest risk level. Detailed description of these groups is provided below: Group evaluation of technical- technological IP risk is composed of: qualified labour force, supply of construction materials, designing mistakes, course of the constructional works. Group evaluation of constructional IP risk is composed of: transport problems, supply problems, production quality, management quality. Group evaluation of political IP risk separate criterion. Group evaluation of financial IP risk is composed of: insolvency situations during construction, decrease of project production price in the market, construction expenditure, fluctuations in resource prices. Group evaluation of ecological IP risk is composed of: accidents, laws on environmental requirements, change in the management attitude towards the project. Group evaluation of legal IP risk is composed of: failure to comply with the contracts, inaccurate construction documentation, failure to harmonize the laws, internal and external legal processes. Further the classification of the possible investment project risks must be established taking into consideration all levels of their multi-purpose quality descriptions. During that quality of the received results must be checked as well. First of all, classification of the factors described in the second level is established. Quality class common evaluations of the first hierarchy level. After classification these common evaluations are filled with concrete contents. Afterwards classification of the first level factors is provided. Final result rules for solving investment project risk evaluation problem. It is suggested to establish the risk of the constructional investment project employing verbal analysis, using CLARA method, which is based on

classification that allows evaluating constructional investment project by the decision made according to the accurately established classes taking into consideration the respectful offered criteria for risk size evaluation. 3. Classification course. 1 STAGE. Evaluation of technical technological investment project (IP) risk (Fig. 3). For second hierarchy level evaluation criteria are introduced: Criterion 1 qualified labour force; Criterion 2 supply of construction materials; Criterion 3 designing mistakes; Criterion 4 course of the constructional works. Criteria evaluation classes: Class A high; Class B average; Class C low. For example, if such a combination is put into the program: 1. (1) Qualified labour force Average. 2. (1) Supply of construction materials Average. 3. (0) Designing mistakes High. 4. (0) Course of the constructional works High. The expert allots it to class A high evaluation. When the assigning is finished, a transfer is made to the next stage (by pushing the button NEXT ). Another evaluation combination is provided. This is done until all the combinations are allotted to the respectful class. During the work the expert might make a mistake or change his opinion, therefore, contradictions might appear in his answers. In such case, the program shows a warning that contradictions have occurred and it will ask to confirm the new answer or to change it. Fig 4. Evaluation of the alternative Fig 3. Criteria for evaluation of technical-technological investment project risk. Criteria 1 4 are chosen for evaluation of technical technological IP risk. While analysing the project the expert determines whether the chosen labour force is qualified enough, whether permanent continuous supply of materials will be ensured during the construction, what the estimated course of jobs is. After the project is analysed, it is determined if there are no mistakes in it. Others evaluation of investment project risks (financial, ecological and any) data input into the program is analogous to the first stage. Classification is performed after verbal risk evaluation scheme data are put into the program. Classification implementation in the program. After introducing all the criteria that will be taken into consideration to classify all available investment projects, the process of classification can start. The classification (Fig. 4) is made in the following way: the program composes combination of the criteria evaluations. The expert assigns the evaluations combination to the respectful class. If program CLARA is used, all the contradictions are eliminated during the work. Fig 5. Evaluation of the alternative After the work is finished, the program saves all the data, performs analysis and shows the number of the given DM questions, the number of classified combinations and the number of eliminated combinations. It also shows how many of the evaluated combinations were assigned to classes A, B or C. Evaluations of all

second hierarchy level criteria are established in an analogous way (Fig 5). 7. Conclusions. 1. In practice it is impossible to avoid not exhaustive and inaccurate information, therefore, unfavorable risky situations occur, the consequences of which can be very damaging to the project. Due to close cooperation of the participants of the project the risk occurring in one stage of the project can transfer to other stages and one type of risk can change into another one. This means that chain reaction is characteristic to the risk and it decreases efficiency and safety of any project. 2. In the present paper, a novel algorithm CLARA is offered for ordering multicriteria alternatives. It differs from the existing similar methods in the wider range of application allowing it to be used with various scales of criteria evaluation, a random number of the solution classes, incomplete order on the criteria scales as well as in the considerably rarefied space of the alternatives. 3. The suggested algorithm is more effective in terms of the time spent by an expert. A comparative calculation of the efficiency of the algorithms used in classifying the objects in the order of their significance has shown that CLARA is much more effective than DIFKLASS, CLANSH and other algorithms. 4. Investment risk in construction can be evaluated efficiently enough using CLARA method. This method allows to classify all possible constructional investment projects presented by evaluations on the predefined criteria into several accurately defined classes reflecting the project risk level. A method contains an algorithm to achieve the minimal amount of the DM questions. References 1. Ustinovichius, L.; Zavadskas, E. K.: Assessment of investment profitability in construction from technological perspectives. VGTU, Vilnius: Publishing house Technika, (2004) (in Lithuanian). 2. Ashihmin, I. V., Furems, E. M.: UNICOMBOS - Intellectual Decision Support System For Multicriteria Comparison And Choice. Artificial Intelligence (Искуственный интелект) Proceedings of the Ukrainian Academy of sciences, Kiev, 2 (2004) 243 247 (in Russian). 3. Korhonen, P., Larichev, O., Moshkovich, H., Mechitov, A., Wallenius, J.: Choice behavior in a Computer-Aided Multiattribute Decision Task. J. Multicriteria Decision Analysis, Vol. 2, No 2(2), (1997) 43 55. 4. Larichev, O., Kochin, D., Kortnev, A.: Decision support system for classification of a finite set of multicriteria alternatives. Decision Support Systems, Vol. 33, No 1 (2002) 13-21. 5. Ustinovichius L., Kochin D.: Verbal analysis of the investment risk in construction. Journal of Business Economics and Management, Vol. 4(4), (2003) 228-234. 6. Larichev, O., Mechitov, A., Moshovich, E., Furems, E.: Revealing of expert knowledge. Publishing house Nauka, Moscow 1989. (in Russian). 7. Larichev, О. and Bolotov, А.: System DIFKLASS: construction of full and consistent bases of expert knowledge in problems of differential classification. The scientific and technical information, a series 2, Informacionie Procesi i Sistemi (Information processes and systems) (9) 1996 (in Russian). 8. Asanov, A., Borisenkov, P., Larichev, O., Nariznij, E., Rozejnzon, G.: Method of multicriteria classification CYCLE and its application for the analysis of the credit risk. Economy and mathematical methods (Ekonomika i Matematicheskije metodi), Vol. 37(2), (2001) 14-21 (in Russian). 9. L. Ustinovichius, G. Ševčenko, D. Kochin.: Classification of Real Alternatives and its Application to the Investment Risk in Construction. In book: Multiple criteria decision making 05. Edited by T. Trzaskalik, Katowice (2006) 299 318. 10. Zavadskas E. K., Ustinovichius L., Stasiulionis A. Multicriteria valuation of commercial construction projects for investment purposes. Journal of Civil Engineering and Management, Vol. 10, 2, 2004, p. 151 166. 11. Kochin D. Yu. Building expert knowledge bases to be used for intellectual training systems. Doctoral thesis, М.: ISA RAN, 2006. 12. Nedzveckas J., Rasimavichius G. Investment currency risk and ways of its decrease. Economy. Publishing house of VU, Vilnius. Vol. 51, 2 (2000) 63-74. 13. Naryzhny Y. V. (1998). The development of intellectual training systems based on expert knowledge. Doctoral thesis, М.: ISA RAN, 1998.