Stock Price Trend Prediction using Artificial Neural Network Techniques

This paper presents a predictive model which to predict the trends of stock prices using Data Mining techniques. This research will allow the investor to make a more informed decision to buy and sell stocks, and in the most appropriate period. The predictive concept in this work implies learning historical price patterns, indicators, and behavior; and then predicting the future trends in one, five, and ten day periods. We compare the effectiveness of feature selection using Gain Ratio Attribute with the Ranker Search Method and Wrapper Selection with Greedy Step Wise Search Methods. Interestingly, we can reduce the attributes from 14 to 6 (57.14%) using Wrapper Subset Evaluation with Greedy algorithm through forward selection. Accuracy improved over the models which were built from the original number of attributes. The results of our experiment demonstrate that the predictive model for weekly (5 and 10 days) stock price direction is improved through the use of Artificial Neural Network (ANN) classification, in which the maximum accuracy of the model reached 93.89% at 10 days prediction, which were a vast improvement to the daily and 5 day predictions employing only six selected input attributes.