Person movement prediction using artificial neural networks with dynamic training on a fixed-size training data set

Significant technical development over the last years has lately been showing more and more promise of making the vision of smart environments come true. The role of future smart environments lies in proactive interaction. Prediction of user’s actions plays a vital role in such interaction. This paper presents a method based on artificial neural networks designed to accommodate the problem of person movement prediction. The paper explores the importance of dynamic training in prediction of nonstationary time series. An approach to dynamic training, based on the so-called on-the-fly training, is presented.