Sentiment-based support vector machine optimized by metaheuristic algorithms for cryptocurrency forecasting

Abstract

Time series are used to model a variety of financial phenomena. The cryptocurrency forecasting problem is the focus of this thesis, which investigates time series forecasting challenges in finance. However, earlier research has neglected to consider the importance of sentiment and public opinion in today's market. The Commodity Channel Index (CCI), historical data and a machine learning algorithm are also employed in this study to improve the accuracy of time series forecasting. By employing hyperparameter optimization, this thesis intends to offer a novel sentiment-based support vector machine optimised by particle swarm and moth-flame optimization algorithms (SVMPSOMFO). PSO, GA, WOA, GOA, GWO, HS and MFO are compared against the proposed algorithm's performance for predicting cryptocurrency prices. A thorough investigation and discussion of all experimental results on different datasets are performed. From the findings, SVMPSOMFO outperforms other optimization methods in terms of accuracy rate when compared to a prediction model that excludes sentiment information. In addition, statistical tests are performed to validate the outcomes of the study.