Research On Uncertain Event Prediction Based On SVM

Recently, with the widespread use and rapid development of current EDGEs (Electronic Data Gathering Equipments), for example, RFID (Radio Frequency Identification), unprecedented abundance of dynamic event streams are emerged. Event processing technique has recieved abroad attention and research for the past few years, as it can help people to acquire useful message from the massive amounts of data. Event processing is widely applied in commercial inspection and prediction, supply chain management, climate and environment supervision and medicial field, etc..,but most research is based on detection for certain event data at present. It can do nothing with probabilistic flow which is formed by uncertain events. On the contrary, we need to infer the latter event or event that will happen in a certain time in the future by studying the former events. So far, there is no research about the probabilistic flow prediction. It is urgent to do some research on uncertain event prediction.In order to realize uncertain event prediction, this paper has designed a special event storage structure, data grid, combining the theory of traditional Support Vector Machine (SVM) and the characteristics of probabilistic flow, and makes the uncertain event prediction possible by bi-directional modeling and two-step modeling strateges which is based on data grid. In this paper, the main contributions are as follows:First, it proposes a prediction method for uncertain events by bi-directional modeling and two-step modeling based on SVM.Second, it proposes the event storage structure, data grid. The structure can support data cleaning and filling and provide horizontal and vertical data. Besides, it can also offer the best train data for building prediction model.Finally, it designs updating strategies for horizontal and vertical models based on the grid structure, which can quickly set up a new accurate prediction model.Experiments show that a series of schemes proposed by the thesis, including data preprocessing, event storage, modeling, prediction and model updating strategy, can predict the uncertain event and have high performance and accuracy of the prediction.