Financial Market Prediction With Hybrid CNN-LSTM Models In Business Management

Abstract

Predicting trends in the financial market has been known to be quite difficult, considering the dynamics and non-linearity of the financial time series dataset. Statistical models like ARIMA and linear regression fail to model complex relationships or learn any sort of dependency within the data to predict trends. In order to deal with the problems of predicting the trends of financial markets, this research work offers a Hybrid CNN-LSTM architecture that makes use of both the CNN and LSTM architecture models in order to forecast future financial market values. Data from the historical stock markets and other economic factors were gathered and pre-processed. This model was implemented using Python, TensorFlow, and Keras libraries. The accuracy, precision, recall, F1-Score, and AUC obtained from the experiments reached the following high numbers, respectively: 88%, 0.89, 0.87, 0.91, and 0.95. Statistical test results revealed a high level of significance, and the confidence interval of [87%, 89%] with a standard deviation of 0.02 further proves the consistency of the model. The experiments and control tests proved that employing CNN, LSTM, and dropout layers increased the accuracy of predictions significantly. The proposed hybrid CNN-LSTM model can serve as a forecasting model and help investors analyze the market trend, but it should not rely on the prediction results for investment decisions or assessing their risk.