A Swarm Optimised Deep Learning Model for Financial Time Series Forecasting
Abstract
Prediction of data on financial time series is extremely difficult due to the inherent complexity and constantly changing dynamics of financial market. Deep learning (DL) methods are found efficient in capturing such underlying dynamism. This article uses a Particle Swarm Optimisation (PSO) to fine-tune the biases and weights of a popular DL method, i.e., Long Short-Term Memory (LSTM) network which resulted in the creation of the ground-breaking hybrid model LSTM+PSO. To evaluate LSTM+PSO, it is used to forecast the values of two widely tracked currency exchange rates. Our concurrent use of the conventional LSTM for the same prediction task serves as a benchmark for comparison. The MSE statistic is used for performance evaluation. According to the results, LSTM+PSO significantly outperform basic LSTM in terms of making correct predictions. Exhaustive experimental outcomes show that the combined forecast is good at dealing with the complex problems of currency exchange rate time series forecasting.
Authors: Sudersan Behera, Sarat Chandra Nayak, Sanjib Kumar Nayak, Sung-Bae Cho
Published in: CRC Press eBooks (2024)