Considering factors affecting the prediction of time series by improving sine-cosine algorithm for selecting the best samples in neural network multiple training model

Abstract

As stock exchange is complex and there is a high volume of information to process, no good prediction results are obtained using a simple system. Therefore, researchers have presented combined model to propose a system with lower sophistication and higher accuracy. System only uses information of one index for predicting in most prediction models but a two-level system of multi-layer perceptron neural network is proposed in this model and several indices are used to prediction and sine-cosine algorithm is also used to select the best samples after neural network training in order to train the neural network better and consequently gain better results. Results show that the proposed model is able to perform with lower prediction error compared with other models.