Neural multivariate prediction using even-knowledge and selective presentation learning

We investigate ways to use knowledge and network learning techniques to improve neural multivariate prediction ability. The prediction of daily stock prices was taken as an example of a complicated real-world problem. We make use of prior knowledge of stock price predictions and newspaper information on domestic and foreign events. Event-knowledge is extracted from newspaper headlines according to prior knowledge. We choose several economic indicators according to prior knowledge and input them together with event-knowledge into neural networks. Also used is a selective presentation learning technique for improving the ability to predict large changes by neural networks. We present training data that correspond to large changes in the prediction-target time series more often than those corresponding to small changes. The effectiveness of our approach is shown experimentally