The inter-dependency in evaluating and selecting portfolio of multiple items means that concurrent or subsequent choices are dependent on the fact that the most prefered items have already been selected. There are previous studies on "behaves in the evaluation and selection of the inter-dependent choice of portfolio of multiple items" in marketing area. These studies have common thing that they address the subject by developing multiattribute preference model, which can be various by the method that treats the interaction between the items consisting the portfolio, and that the use the nortion of satiation in terms of each attribute or not. In this paper, we attempt to provide a new tool for predicting consumer behavior by adopting neural network approach. In order to carry out the experiment, we adopt the credit portfolio selection behavior which is one of the most suitable items to the Lancasterian tenet. For a comparative analysis of the predictability, the hit ratio of the proposed neural network is compared with those of previous models, the attribute satiation model, the flexible attribute overlap vector model and the fixed attribute overlap vector model. The result show that the proposed model is superior to the previous models in the prediction power and relatively easier to apply in practice than the previous models, for being free from the load to adopt what scheme of interation among items making up a portfolio and being able to save the cost to attain the additional information of attribute overlap.
다품목군(Portfolio of multiple items)의 평가와 선택에 있어 존재하는 상호의존성은 이러한 행위를 예측하는 모형을 어렵게 만들고 있다. 즉 품목들 간의 상호작용과 포만성(Satiation)을 어떻게 고려하는가가 모형을 구분하는 척도가 되고 있다. 본 연구는 이러한 기존의 방법을 벗어나 인공 신경망(Artificial Neural Network) 기법을 이용한 다품목군의 선택 행위 예측 모형을 제시한다. 실험을 위하여 크레디트 카드 포트폴리오의 선택행위에 대한 설문 조사를 실시하였고, 이 자료를 이용하여 예측력을 실험하였다. 예측력의 비교을 위하여 기존의 모형 세가지를 채택하였다. (the Attribute Satiation model, the Flexible Attribute Overlap Vector model and the Fixed Attribute Overlap Vector model). 인공 신경망 기법을 이용한 본 연구의 모형은 세가지의 구성이 채택되었는데, 이들 모두 예측력에 있어서 기존의 모형들을 능가하였으며, 속성들간의 상호작용이나 중복 효과(Overlap Effect)를 고려하지 않아도 예측력에 있어 우수한 결과를 나타내, 이들을 얻기위한 모형구성상의 노력이나 정보를 얻기위한 노력을 감소시켜 주는 것으로 나타났다.