Long short-term memory algorithm for Stock Market prediction | #finalyearprojects 2020

Abstract— In this project, deep learning will be used to predict the stock price movement and machine learning will going to use for sentiment analysis. Scoring provided for sentiment analysis in order to observe the relationship between the stock market and the news headline score. About the prediction part, LSTM is going to be the prediction model because it is good in memories the past information and able to eliminate irrelevant data. Input dataset obtained from SHAREINVESTOR.com because it contain the history data up to 20 years ago. There are 2 types of datasets are downloaded from the website which are historical price dataset and intraday price dataset. Historical price dataset is used for long term prediction (few months) while the intraday price dataset is for short term prediction (few days). The downloaded dataset will be processed normalization by using min-max normalize technique to increase the accuracy of forecast price. Next, normalize data will going to be trained by stacked LSTM model which comprised of multiple LSTM layers where each layer contains multiple memory cells. In addition, to validate the prediction data, different metrics are chosen as our validation method for forecast data. Currently the prediction model accuracy is improved up to 55% which indicated that the predictive effect of LSTM could reach 14% better than traditional FFNN prediction model. Apart from that, sentiment analysis going to analyze the news headline which are provided from News Straits Times official website. Furthermore, it will show the polarity in word format such as positive, negative or neutral to indicate the polarity of the news headline.
#fyp #stockmarket #predictions
Contact supervisor: Dr Ahmad Shahidan Bin Abdullah
Department of / Centre of
Faculty of Electrical Engineering
Universiti Teknologi Malaysia
Johor Bahru, Malaysia.