Financial time series prediction using machine learning: methods, models, and performance

The prediction of financial time series is inherently difficult due to nature of financial data, which is non-stationary, random, and constantly evolving. In this paper, the performance of distributed machine learning models, to be specific functional link artificial neural networks (FLANN), is evaluated using incremental as well as diffusion-based learning. In the context of exchange rates, stock indices and Net Asset Value (NAVs), the proposed research establishes that the methods of this study are superior to the conventional methods in terms of predictive ability and computational overhead. The incremental model is effective in modeling short-term movement while the diffusion model is effective in moderating long-term fluctuations. As the data shows, distributed learning requires less computation and enhances the option for scalability – making it a feasible solution for distributed financial data analysis. This work can be classified under the domain of applying machine learning in finance and gives knowledge on the more productive way of handling big data.