Application Of Deep Learning In Financial Time Series Classification

Time series data has the characteristics of temporal correlation,mass and high dimensions,which makes time series classification a challenging problem in data mining.With the emergence of new types of network architecture,deep learning has achieved very significant results in the fields of computer vision and speech recognition.At the same time,deep learning has also been shown to have considerable empirical effects on time series classification.Financial time series classification refers to the classification and decision making of financial time series data.The analysis and prediction of financial time series data is the key to constructing profitable quantitative trading investment strategies.As a complex,non-linear dynamic system,financial market classification and forecasting is a challenging and practical research direction in the financial field.This research is to explore the effect of the new deep learning architecture applied to financial time series classification,and the profitability of constructing quantitative stock selection trading strategies.At the same time,it compares and analyzes how different feature expression forms of financial time series affects the differences in the learning effects of corresponding types of deep models.Since feature expression in different data forms will have an impact on the learning effect of the deep model,this research first performs visual feature engineering on time series,specifically uses GADF algorithm and candlestick chart visualization of financial data to convert time series data into image data.Based on various data forms and combining data characteristics,deep learning models are constructed for comparative analysis.The models specifically include the LSTM model for time series classification,the CNN model for GAF images recognition,and the Conv LSTM model for candlesticks images recognition.Then a quantitative trading backtesting framework is built,generating stock selection portfolios based on the training results of the model,and backtesting is performed based on trading signals.The performance of each model is compared and analyzed from multiple sets of transaction evaluation indicators of backtest curves.The experimental results in this paper show that the application of deep learning to construct quantitative stock selection trading strategies has good profitability and risk control capabilities.At the same time,the overall performance of the image classification models CNN and Conv LSTM based on visual images is significantly better than the time series classification LSTM model.