Applying a hybrid data-mining approach to prediction problems: a case of preferred suppliers prediction

Data mining has emerged as a powerful tool to automate the decision-making processes. The deficiency of the traditional approach of data mining for prediction is that the low-accuracy decision rules that cannot be used and the quality rules, with high accuracy, do not guarantee that they will be used since the premise (condition) part of the rule should match with the input domain of the testing data sets. Furthermore, negative information/data, which are defined as information/data, misclassify the outcomes from the testing data set and are normally discarded. It is possible to reuse the information/data for training purposes and they contain a positive impact on prediction accuracy. This paper presents a data-mining-based hybrid approach that consists of a novel rough-set algorithm for feature selection and an enhanced multi-class support vector machines (SVMs) method for accurate prediction. The approach can simultaneously derive decision rules, identify the most significant features and generate a well-tuned prediction model with high accuracy. A case study of supplier selection of a video game system is validated by historical data and the results show the practical viability of the hybrid approach for predicting preferred suppliers.