Financial Forecasting in Volatile Markets Using Hybrid LSTM and Genetic Algorithms

Abstract

In volatile markets, forecasting becomes a daunting task since market conditions are very unpredictable. At times when there are good or bad things, it all depends on the economic or political events occurring at a particular period, and also the sentiments of the investors. All the forecasting methods that use ARIMA and SVM have generally poor performance in predicting trends in highly volatile markets. The intention behind the creation of this research paper is to develop a hybrid approach by integrating the LSTM network’s sequence modeling capabilities with the Genetic Algorithm (GA) for feature selection and optimization. The main aim of the study is to enhance financial forecasting models through improving their performance and reliability. The process consists of three main stages, including the pre-processing of financial data, feature selection by applying the GA approach, and model development using the LSTM neural network. It is clear that the Hybrid LSTM-GA model performs better than others by showing better accuracy and precision, as well as having a good value for its MSE and RMSE measures. Moreover, the model shows its flexibility, since it performed efficiently both under high (91.2%) and low (93.8%) volatility conditions. In summary, the Hybrid LSTM-GA model proves more effective than other models in predicting outcomes since it can handle uncertainties and changes in data. Therefore, it can be applied in stock market analysis and forecasts. It can effectively predict results when there is high uncertainty in the stock market environment.