Stock turning point recognition using multiple model algorithm with multiple types of features

Stock turning point has been playing a significant role in the stock investment as buying or selling stocks around it can make a good return. The existing methods recognize the turning points according to different types of features but the effect is not so satisfactory. In this paper, we firstly build multiple types of features what reflect many aspects of characteristics of stock and utilize mutual information to partition them into several subsets. Then we use SVM to train the turning point recognition model on every feature subset. Finally, we combine the multiple models into an ensemble and apply Particle Swarm Optimization (PSO) to optimize the combination coefficients. Experimental results show that our method is more effective.