StockPrediction_Project using GAN

1. Dataset

• Using dataset since 2000 in FAANG (FACEBOOK, Amazon, Apple, Netflix, Google) dataset
• I think it shows the stock data of IT companies growing in the stock market well.
• https://www.kaggle.com/aayushmishra1512/faang-complete-stock-data?select=Apple.csv

2. Project goals

• (1) Soving follw up problem(right shift) with LSTM model in stock prediction. This problem is relevent to information vanishing with LSTM
• (2) Making model that predict whether tomorrow stock' price will be increased or decreased
• Therefore, Create a predictive model for Adj Close using stock data from 2000 to 2020 of FAANG companies

3. Analyzing previous approaches

• Bidirectional LSTM is consisted of 20 LSTM cell in hiddem layer and 1 output in linear layer.
• Bidirectional LSTM uses RMSE in objective and Adam in optimizer.
• reference paper : ‘양방향 LSTM 순환신경망 기반 주가 예측 모델’
• paper link : https://scienceon.kisti.re.kr/srch/selectPORSrchArticle.do?cn=JAKO201819355173173&dbt=NART

4. Time Series GAN

• reference paper: Stock Market Prediction on High-Frequency Data Using Generative Adversarial Nets
• link : https://www.hindawi.com/journals/mpe/2018/4907423/
• LSTM in the generator predicts the data of the next time point like the basic LSTM model.
• In Discriminator, the existing stock price data from time 1 to time T and the generated prediction value were received as input to distinguish the authenticity of the data. 1d convolution layer was used in discriminator

5. Experiment result

Comparing prediction graph

(1)Predict test data in increase pattern

(2)Predict test data in decrease pattern

Comparing evalation metric