The paper investigates methods of artificial intelligence in the prognostication and analysis of financial data time series. It is uncovered that scholars and practitioners face some difficulties in modelling complex system such as the stock market because it is nonlinear, chaotic, multi- dimensional, and spatial in nature, making forecasting a complex process. Models estimating nonstationary financial time series may include noise and errors. The relationship between the input and output parameters of the models is essentially non-linear, where stock prices include higher-level variables, which complicates stock market modeling and forecasting. It is also revealed that financial time series are multidimensional and they are influenced by many factors, such as economics, politics, environment and so on. Analysis and evaluation of multi- dimensional systems and their forecasting should be carried out by machine learning models.
The problem of forecasting the stock market and obtaining quality forecasts is an urgent task, and the methods and models of machine learning should be the main mathematical tools in solving the above problems. First, we proposed to use self-organizing map, which is used to visualize multidimensional data by configuring neurons to quantize or cluster the input space in the topological structure. These characteristics of this algorithm make it attractive in solving many problems, including clustering, especially for forecasting stock prices. In addition, the methods discussed, encourage us to apply this cluster approach to present a different data structure for forecasting. Thus, models of adaptive neuro-fuzzy inference system combine the characteristics of both neural networks and fuzzy logic. Given the fact that the rule of hybrid learning and the theory of logic is a clear advantage of adaptive neuro-fuzzy inference system, which has computational advantages over other methods of parameter identification, we propose a new hybrid algorithm for integrating self-organizing map with adaptive fuzzy inference system to forecast stock index prices. This algorithm is well suited for estimating the relationship between historical prices in stock markets. The proposed hybrid method demonstrated reduced errors and higher overall accuracy.
Comparing classic time series models and the LSTM recurrent neural network: An application to S&P 500 stocks
