Abstract: Three experiments were designed to investigate how the subjects were influenced in the course of making feature prediction in the uncertain circumstance of classifying when the subject and the feature were associated or separated Experiment 1 explored whether the subjects used the non-target categories in the course of making feature prediction when the base probability of the feature within the non-target categories was increased The results in Experiment 1a and 1b show that, even though the base probability of feature prediction for non-target categories is enhanced, subjects only consider the target category in their feature probability prediction without considering the information from other non-target categories Experiment 2 explored whether the association of the dimensions within the non-target categories promoted the use of the non-target categories The results prove that, in the case of the association of the two dimensions in non-target categories, the subjects will use the non-target categories information when they are making a feature prediction, which is in conformity with the Bayesian rule Experiment 3 explored whether the feature prediction was influenced when the rate of association of the dimensions within the target category was promoted The results show that the raise of proportion in the association of target and prediction feature in target category will enhance the feature prediction probability The results of the three experiments showed:(1) If the dimensions within the non-target categories were separated, subjects didn't take into account the non-target categories, following the single-category view (2) When the dimensions were associated within the non-target categories, the subjects would use the information of the non-target categories (3) If the proportion of the association in the dimensions within the target category were raised, the probability of the feature prediction would be enhanced According to these results, the proportion of association of the object and the feature should be thought as an important variable, which should be put into the formula of the BayesianRule

Key words: classification, feature prediction, non-target category, association in dimensions, separation in dimensions