Machine learning methods for financial time series prediction: techniques and challenges
Abstract
Financial markets are volatile and non-stationary which makes it a complex task to predict the markets. Based on the data of 2014-2024, this study investigates the application of statistical models, deep learning and ensemble methods to time series forecasting. Results indicate that hybrid CNN-LSTM models and ensembles are the best, with Random Forest and XGBoost achieving 93.1% accuracy. The paper also highlights preprocessing, hyperparameter optimization and computational performance as key factors to apply the study in a real-world dynamic financial setting.
Fields of Study & Subjects
Stock Market Forecasting Methods
Time Series Analysis and Forecasting
Forecasting Techniques and Applications
Publication details
- Journal / source
- CRC Press eBooks
- Pages
- 475–480
- Publication date
- 17 February 2026
- Language
- EN
- Document type
- Book chapter