Self Organizing Maps (SOM) with Moving Average features for Stock Market Prediction

File Size:
1.05 MB
Volume 3, Issue 9 (September, 2017)
Publication No:
Supreet Kaur, Mr. Sachin Majithia
7 x

The proposed model has been entirely based upon the guided trading data extraction using the historical data aware SOM (Self Organizing maps) with multilayer perceptron algorithm along for the stock price prediction which is better than the existing Multi Classifier ensemble (MCS) that uses bagging and boosting. This model has been designed to work over the live stock market data obtained from the Google Finance API for the selective stocks in the selective trading sectors. It has been tested with the downloaded historical stock price data for six years, which has been prepared in the training data according to different window sizes with variable number entries obtained from continuous series of trading prices. In this paper, it has been tested for the various performance measures which includes the predictive accuracy and F1-measures. The proposed model has been found accurate higher than 90% in all of the rounds if the true negative cases are also being analyzed. This model has been recorded with the average accuracy over all of the test cases nearly at 93% which is better all of the other models used under the existing model of accuracy 88.2%. The proposed model has outperformed all of the existing models designed with the different filters over the differently processed datasets

Stock price prediction, Future trend analysis, SOM, Multilayer perceptron, Neural network, predictive analysis.