Abstrait

Stock price prediction model based on modified IWO neural network and its applications

Hong Liu, Xiaoyan Lv

For the deficiency of easily trapping into a local optimum of neural network learning algorithm based on steepest descent method, we apply a new kind of optimization strategy called invasive weed optimization (IWO)algorithm, where the standard deviation of the distribution of offspring individuals adjusts dynamically, and a feed-forward neural network learning algorithm based on IWO is given. Considering the phenomena of optimization accuracy being not high and precocious in IWO algorithm, further we use IWO algorithm to search globally, meanwhile searching locally by utilizing reflection, extension and compression operation of complex to create a new solution for replacing the worst individual in current population, and propose a complex invasive weed optimization (CIWO) algorithm, which avoids precocious phenomenon, improves precision of optimization and increases the speed of convergence. On this basis, neural network prediction models based on IWO and CIWO algorithm are respectively established, the feasibility of IWO algorithm training the neural network and the validity of improving the IWO algorithm with complex method are verified by stock price forecasting.