Recently according to the scale of development projects in the field of construction becomes bigger and consumers’ demands are diversified, uncertainty of development projects is being increased as time goes by. It can be operated to be a factor to ...
Recently according to the scale of development projects in the field of construction becomes bigger and consumers’ demands are diversified, uncertainty of development projects is being increased as time goes by. It can be operated to be a factor to increase a risk in the process of predicting and managing cost of construction from early stage of a project.
Therefore, it is important to predict const of construction with limited information. To predict cost of construction of a new development project because of features of construction projects, it is necessary to analyze data of construction expenses of the existing performed construction work. The study could see that case based inference with experience -oriented problem-solving technique was the most suitable to predict cost of construction. However, even if cases similar to a new case through case based inference is searched, there are no cases reflecting construction features of a new case in the same way. Therefore, it was suggested that it was necessary to study on correcting construction expenses of inquired similar cases according to construction features of a new case.
To improve forecast accuracy of early construction expenses in development projects of apartment house, the study suggested a plan to correct construction expenses of searched similar cases in the stage of case based inference with genetic algorism. The purpose of the study was to improve forecast accuracy of searched construction expenses through case based inference. The results of the study were as follows. (1) The study considered preceding researches related to prediction of construction expenses of apartment house and judged that it was necessary to select influencing factor to improve forecast accuracy of early construction expenses and to improve convenience of DB construction and efficiency of forecast process.
The study selected 3 nominal factors and 12 quantitative factors to be an influencing factor of the final construction expenses of apartment house by carrying out interviews with experts for selection and verification of influencing factor in cost of construction.
(2) As a result of multiple regression analysis, with backward elimination approach, the study calculated roof form (0.126), total floor area (0.848), site area (0.374), building area (0.464), average unit housing area (0.118) and story height (0.116) and induced weights of total floor area (0.97), finishing level (0.09) and building coverage ratio (0.09). Therefore, it is judged that searched forecast accuracy will be increased more than when using single variables of total floor area by including variables besides total floor area in forecast process.
(3) When the study inquired similar cases by applying attribute weights calculated by backward elimination approach, the average error rate was 17.51% and was 12.07% when advance• stage selection approach was applied. However, even if a case was selected to be a case similar to a new case, there were differences between attribute value of the new case and that of the similar case. Therefore, it is judged that it is necessary to modify construction expenses of inquired similar case since it is difficult to apply the new case to the similar case right now.
(4) When the study modified construction expenses searched by genetic algorism, in the event of searching it through backward elimination approach, the average value of error rate became lower from 17.51% to 6.90%.
Therefore, it was verified that it was efficient to modify construction expenses of similar cases with genetic algorism. Even if case database was enough constructed, it is judged that it was possible to increase the utility of forecast in construction expenses by applying the above process.
(5) When attribute weights calculated by advance• stage selection were used, the average error rate was reduced from 12.07% to 6.32%. It had lower error rate than when using weights calculated by backward elimination. However, the error rate of two cases out of 10 verified cases was increased. Therefore, it had somewhat low reliability. It is judged that it is necessary to make more prudent judgment when the number of influencing factors is small since the number of influencing factors is small.