The prediction of rare events is a challenging scientific problem. Events such as extreme meteorological conditions may aggravate human morbidity and mortality. Yet, their prediction is inherently difficult as, these events are characterized by low occurrence, high sampling variation, and uncertainty, e.g, earthquakes have a high magnitude variation and are irregular. In the past, attempts have been made to predict rare events using linear time series forecasting algorithms, but they failed to capture the surprise events.
This study proposes a novel strategy that extends polynomial neural network (PNN) techniques. As a new strategy, we retain and reuse error feedback plus other relevant information, as part of the multivariate sample data that provides relevant multivariate inputs to the PNN. This is important because the strength of PNN, is in predicting outcomes from multivariate data, and it is very noise-tolerant and capable of modeling highly non-linear relations, like those in rare events.
In this project, we innovate and apply new rare event prediction algorithms into case-studies of spatial wireless sensor networks. E.g. WSN for environmental monitoring (from land, forestry, to water). The algorithms are designed to be efficient and possible for real-time monitoring integration.
The prime objective is to innovate new rare-event forecasting algorithms; secondary objective is to integrate into real-time monitoring systems. Innovation is to be achieved by four means: –
(1) Innovate a new breed of rare-event forecasting algorithms, possibly from the kernels of incremental data mining algorithms;
(2) Extend from polynomial neural network methods with the aim of achieving low error-rates; we have already achieved certain preliminary results in past year research; what’s needed is extensive testing with a wide range of data.
(3) Apply these new algorithms in a case study of real-world wireless sensor network data.
(4) So far rare-event prediction such as global earthquake is singular in a sense of predicting “when” a mega-quake may hit us somewhere in the world. In this project, we attempt to extend such prediction with the local context, i.e., we assume events that happen in one location may have relations to others. By considering the spatial context, new algorithms are needed that reach out further hyperspace dimensions in computation.
With the achievement of innovating new rare-event algorithms, especially those that could integrate with real-time disaster monitoring, the expected significance is the intellectual contribution and knowledge contribution to computer science.
Also, after testing and validating the performance of the new algorithms, our next step is to research further into developing some function monitoring systems with enhanced HCI functionalities.