STOCK MARKET MOVEMENT PREDICTION USING CONVOLUTIONAL NEURAL NETWORK (CNN)

• 主要作者: Jhouma Parulian Napitu, Yohanes
• 格式: Final Project
• 語言: Indonesia
• 在線閱讀: https://digilib.itb.ac.id/gdl/view/54358
• 機構: Institut Teknologi Bandung

In predicting capital market movements, the main concern is determining when is the right time to buy, sell, or hold a stock. Currently, most of the analysis performed by an investor is still manual, especially retail investors, so it is difficult to determine an efficient investment strategy. This final project aims to produce a model that is able to predict future stock movements of a company. This can help investors to determine investment strategies more efficiently. In this final project, the proposed solution is divided into two components, namely the preprocessing component and the prediction component. The preprocessing carried out is converting historical price data into a matrix that represents a candle chart. To predict stock price movements the model used is the Convolutional Neural Network for time-series. To determine the parameters used, an experiment was carried out by retraining based on four years of data. The experimental results show that the selected parameters for the models used are 5, 5, 5, and 5 for the size of the first filter, first pooling, second filtering and second pooling, respectively. The model with selected parameters has an average accuracy of 54.35% and an average area under ROC 0.5576. Unfortunately, this performance is not better than previous studies which predict the same thing with different models, namely an average accuracy of 64.14%.