Wavelet Transform LSTM과 Multi-Head Attention의 결합모형을 이용한 주가 지수 예측

In this paper, we proposed a new deep learning forecasting model named WMA-LSTM (Wavelet Transform Multi-Head Attention LSTM) model that combines Wavelet Transform LSTM and Multi-Head Attention, The Wavelet Transform is used to denoise the input values, and LSTM and Multi-Head Attention are deep learning model for forecasting the time-series data. The market price, high price, low price, closing price, and trading volume data are used to forecast the next day’s price of the stock index in Korea, USA, and China(KOSPI, S&P500, HangSeng Index). The results show that our proposed model has the best forecasting performance among the seven comparative models(ARIMA, W-ARIMA, LSTM, A-LSTM, MA-LSTM, W-LSTM, and WA-LSTM) in all three markets. As a result, combining Wavelet Transform LSTM with Multi-Head Attention to increase stability improves forecasting performance.