Stock trends prediction based on hypergraph modeling clustering algorithm

In this paper, we use hypergraph model to predict the stock trends. Sometimes, it is hard to know what a certain stock will do tomorrow. But here we make each stock as a vertex of a hypergraph and the relations, going up or down of stocks, are abstracted as hyperedges. If some stocks have the same actions, going up or down together, during k days, we can build a hyperedge among these stocks. According to the actions of stocks in the last several months, the hypergraph for these stocks is established. Furthermore, the problem of stocks trends prediction is translated into the clustering of the hypergraph. The stocks in the same cluster set always have the same trends. Then the trend of one stock can be predicted by the rest stocks in the same set. And, of course, what the stock did during the last several days should be also taken into account for the prediction. In order to get an accurate prediction, we set a new optimization index for the segmentation of a hypergraph, which is called modularity. Through the definition of modularity, it is not necessary for us to preset the partitioning number any more. When the modularity is immediately below the critical value, the partitioning is stopped automatically. Experiment result shows that our proposed scheme achieves fine stock trend prediction.