Artificial neural networks with feature transformation based on domain knowledge for the prediction of stock index futures

A feature transformation method based on domain knowledge for artificial neural networks (ANNs) is proposed. The method of feature transformation based on domain knowledge converts continuous values into discrete values in accordance with the knowledge of experts in specific application domains. This approach effectively filters data, trains the classifier, and extracts the rules from the classifier. In addition, it reduces the dimensionality of the feature space, which not only decreases the cost and time in the operation but also enhances the generalizability of the classifier. The experimental results of the proposed approach will be compared and tested statistically with the results of the linear transformation method. The results show that the method of feature transformation based on domain knowledge outperforms the linear transformation in modelling of ANNs.