A new ensemble method for gold mining problems: Predicting technology transfer

Of the many available innovative e-commerce technologies, only a small number have been successful in practice. Choosing and purchasing the right e-commerce technology is similar to finding gold in the mountains: there is a low frequency of a desirable state and a high frequency of an undesirable state. Thus, such scenarios are called gold mining problems. In such cases, the goal is to increase the probability of accurately predicting the desirable state. However, few prediction methods are sophisticated enough to predict gold mining problem results accurately. Hence, the purpose of this paper is to propose a novel ensemble method dedicated to increasing the probability of accurately predicting desirable states. We develop the vertical boosting with rewarded vote strategy, which generates classifiers for each attribute in a sample. Each classifier then generates individual rules with the assistance of a sensitivity level, to find desirable states. The individual rule sets are generated with adjustment by the multiplier, and then used in the ensemble method to generate combined rules. To show the method’s soundness, we perform an experiment with a representative gold mining problem: prediction of transferability of the intellectual properties of e-transaction technology.

► Few prediction methods are sophisticated to gold mining problems.
► The vertical boosting with rewarded vote strategy generates classifiers for each attribute in a sample.
► Experiment is performed using real data sets of intellectual property and its transferability.
► The proposed method significantly outperforms the competing ensemble methods.