PREDICTION ANALYSIS OF THE S&P 500 STOCKS PRICE INDEX USING CONVOLUTIONAL NEURAL NETWORK (CNN) AND LONG SHORT TERM MEMORY (LSTM) METHODS, CASE STUDY FOR UP-TREND AND DOWN-TREND STOCKS.

• Main Author: Fadli Ikhsanudin, Naufal
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/57515
• Institution: Institut Teknologi Bandung

This research contains stock price prediction for S&P 500 market using Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM) methods for up-trend and down-trend stocks, which is influenced by the rapid development of neural network in economic field such as stock price prediction especially for CNN and LSTM method. LSTM is one of the Recurrent Neural Network model which has several memory cells and gate units on each neuron whose function is used to manage the memory in each neuron. So this method can easily solve dataset prediction such as time-series dependent data. Moreover, CNN normally recognize and learn for picture and pattern dataset, but also can predict in regression method like stock analyze. Research itself searching on the best parameter for both CNN based on input which are univariate-multivariate and LSTM to predict various stock in S&P 500 index which have up-trend and down-trend dataset and measure the RMSE, MSE, and MAPE every epochs, also the correlative coefficient for every method, whose those values become subject for comparison to know which one the method has the best prediction. The datas have taken from 2 up-trend stocks and 2 down-trend stocks from S&P 500 in certain period of time with various background company. These stock data then analyze with technical apporach to calculate predicted dataset including open, high, low, and close price. The results are for up-trend stock has the best RMSE, MSE, MAPE, and correlative coefficent if approach with LSTM method, which are 71,315 for RMSE, 5085,87 for MSE, 0,01703 for MAPE and also 0,883 for correlative coefficient. Then, for down-trend stock price, has the best error and correlative coefficent if the datas are analyzed with multivariate CNN, which are 0,599 for RMSE, 0,35974 for MSE, 0,0138 for MAPE and also 0,98909 for correlative coefficient. For the future result, this method itself will be aproached with larger epochs and neurons to make sure the values have more accuracy.