Stock market prediction in Bangladesh perspective using artificial neural network

Stock market price prediction is now a prominent and significant issue in financial and academic studies as the stock market plays a vital role in the economy. The process of attempting to anticipate the future valuation of a company's share is known as stock market price prediction. Share prices are time-series information, and artificial neural networks (ANNs) can uncover non-linear associations among time-series information. This makes ANN the best method for predicting stock market values. Many researchers are working on this topic and trying to find the best algorithm which is suitable for predicting the stock price. But significant improvement in prediction is still not achieved. Therefore, in this work, an ANN model is proposed and implemented by using a multilayer feedforward backpropagation method. In this work, data of fifteen companies over a six years span have been analyzed. To predict the specific result, the proposed model has been trained with four different algorithms: Levenberg Marquardt (LM), Bayesian regularization (BR), scaled conjugate gradient (SCG) and Quasi Newton) by changing their parameters. Number of hidden layers, hidden neurons and percentage of training data have been changed to get better output. The split ratio of training, testing and validation data sets is 70:15:15. The projected results are then compared to the actual data after the training and testing procedure to determine the accuracy. The accuracy of LM is 95.64%, BR is 91.26%, SCG accuracy is 88.91% and Quasi Newton is 84.20%. The result showed that, LM algorithm provides better accuracy than other models. In addition, less error has been found from the LM algorithm, making it the best algorithm for prediction in our proposed model.